HAL Id: tel-00838736 https://tel.archives-ouvertes.fr/tel-00838736 Submitted on 26 Jun 2013 HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. High Frequency MEMS Sensor for Aero-acoustic Measurements Zhijian J. Zhou To cite this version: Zhijian J. Zhou. High Frequency MEMS Sensor for Aero-acoustic Measurements. Micro and nan- otechnologies/Microelectronics. Université de Grenoble, 2013. English. tel-00838736
171
Embed
High Frequency MEMS Sensor for Aero-acoustic Measurements
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
HAL Id tel-00838736httpstelarchives-ouvertesfrtel-00838736
Submitted on 26 Jun 2013
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents whether they are pub-lished or not The documents may come fromteaching and research institutions in France orabroad or from public or private research centers
Lrsquoarchive ouverte pluridisciplinaire HAL estdestineacutee au deacutepocirct et agrave la diffusion de documentsscientifiques de niveau recherche publieacutes ou noneacutemanant des eacutetablissements drsquoenseignement et derecherche franccedilais ou eacutetrangers des laboratoirespublics ou priveacutes
High Frequency MEMS Sensor for Aero-acousticMeasurements
Zhijian J Zhou
To cite this versionZhijian J Zhou High Frequency MEMS Sensor for Aero-acoustic Measurements Micro and nan-otechnologiesMicroelectronics Universiteacute de Grenoble 2013 English tel-00838736
THEgraveSE Pour obtenir le grade de
DOCTEUR DE LrsquoUNIVERSITEacute DE GRENOBLE Speacutecialiteacute NANO ELECTRONIQUE NANO TECHNOLOGIES Arrecircteacute ministeacuteriel 7 aoucirct 2006
Et de
DOCTEUR DE THE HONG KONG UNIVERSITY OF SCIENCE AND TECHNOLOGY
Speacutecialiteacute ELECTRONIC AND COMPUTER ENGINEERING Preacutesenteacutee par
laquoZhijian ZHOUraquo Thegravese dirigeacutee par laquoLibor RUFERraquo et
codirigeacutee par laquoMan WONGraquo preacutepareacutee au sein du Laboratoire TIMA
dans lEacutecole Doctorale Electronique Electrotechnique Automatique et
Traitement du Signal
et Electronic and Computer Engineering Department
Microcapteurs de Hautes
Freacutequences pour des Mesures
en Aeacuteroacoustique
Thegravese soutenue publiquement le laquo01212013raquo devant le jury composeacute de
M David COOK Professeur Associeacute Hong Kong University of Science amp Technology Preacutesident M Philippe BLANC-BENON Directeur de Recherche CNRS Ecole Centrale de Lyon Rapporteur
M Philippe COMBETTE Professeur Universiteacute Montpellier II Rapporteur
M Skandar BASROUR Professeur Universiteacute Joseph Fourier Grenoble Examinateur Mme Wenjing YE Professeur Associeacute Hong Kong University of Science amp Technology Examinateur
M Levent YOBAS Professeur Assistant Hong Kong University of Science amp Technology Examinateur M Man WONG Professeur Hong Kong University of Science amp Technology Co-Directeur de thegravese
M Libor RUFER Chercheur Universiteacute Joseph Fourier Grenoble Directeur de thegravese
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
A Thesis Submitted to
The Hong Kong University of Science and Technology
in Partial Fulfillment of the Requirements for
the Degree of Doctor of Philosophy
in the Department of Electronic and Computer Engineering
and
Universiteacute de Grenoble
in Partial Fulfillment of the Requirements for
the Degree of Docteur de lrsquo Universiteacute de Grenoble
in the Ecole Doctorale Electronique Electrotechnique Automatique amp Traitement du Signal
February 2013 Hong Kong
iii
Authorization
I hereby declare that I am the sole author of the thesis
I authorize the Hong Kong University of Science and Technology and Universiteacute de
Grenoble to lend this thesis to other institutions or individuals for the purpose of scholarly
research
I further authorize the Hong Kong University of Science and Technology and Universiteacute
de Grenoble to reproduce the thesis by photocopying or by other means in total or in part at
the request of other institutions or individuals for the purpose of scholarly research
___________________________________________
ZHOU Zhijian
February 2013
iv
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
This is to certify that I have examined the above PhD thesis and have found that it is
complete and satisfactory in all respects and that any and all revisions required by the thesis
examination committee have been made
___________________________________________
Prof Man WONG
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Supervisor
___________________________________________
Prof Libor RUFER
Universiteacute de Grenoble France
Thesis Co-Supervisor
___________________________________________
Prof David COOK
Department of Economics HKUST Hong Kong
Thesis Examination Committee Member (Chairman)
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
THEgraveSE Pour obtenir le grade de
DOCTEUR DE LrsquoUNIVERSITEacute DE GRENOBLE Speacutecialiteacute NANO ELECTRONIQUE NANO TECHNOLOGIES Arrecircteacute ministeacuteriel 7 aoucirct 2006
Et de
DOCTEUR DE THE HONG KONG UNIVERSITY OF SCIENCE AND TECHNOLOGY
Speacutecialiteacute ELECTRONIC AND COMPUTER ENGINEERING Preacutesenteacutee par
laquoZhijian ZHOUraquo Thegravese dirigeacutee par laquoLibor RUFERraquo et
codirigeacutee par laquoMan WONGraquo preacutepareacutee au sein du Laboratoire TIMA
dans lEacutecole Doctorale Electronique Electrotechnique Automatique et
Traitement du Signal
et Electronic and Computer Engineering Department
Microcapteurs de Hautes
Freacutequences pour des Mesures
en Aeacuteroacoustique
Thegravese soutenue publiquement le laquo01212013raquo devant le jury composeacute de
M David COOK Professeur Associeacute Hong Kong University of Science amp Technology Preacutesident M Philippe BLANC-BENON Directeur de Recherche CNRS Ecole Centrale de Lyon Rapporteur
M Philippe COMBETTE Professeur Universiteacute Montpellier II Rapporteur
M Skandar BASROUR Professeur Universiteacute Joseph Fourier Grenoble Examinateur Mme Wenjing YE Professeur Associeacute Hong Kong University of Science amp Technology Examinateur
M Levent YOBAS Professeur Assistant Hong Kong University of Science amp Technology Examinateur M Man WONG Professeur Hong Kong University of Science amp Technology Co-Directeur de thegravese
M Libor RUFER Chercheur Universiteacute Joseph Fourier Grenoble Directeur de thegravese
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
A Thesis Submitted to
The Hong Kong University of Science and Technology
in Partial Fulfillment of the Requirements for
the Degree of Doctor of Philosophy
in the Department of Electronic and Computer Engineering
and
Universiteacute de Grenoble
in Partial Fulfillment of the Requirements for
the Degree of Docteur de lrsquo Universiteacute de Grenoble
in the Ecole Doctorale Electronique Electrotechnique Automatique amp Traitement du Signal
February 2013 Hong Kong
iii
Authorization
I hereby declare that I am the sole author of the thesis
I authorize the Hong Kong University of Science and Technology and Universiteacute de
Grenoble to lend this thesis to other institutions or individuals for the purpose of scholarly
research
I further authorize the Hong Kong University of Science and Technology and Universiteacute
de Grenoble to reproduce the thesis by photocopying or by other means in total or in part at
the request of other institutions or individuals for the purpose of scholarly research
___________________________________________
ZHOU Zhijian
February 2013
iv
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
This is to certify that I have examined the above PhD thesis and have found that it is
complete and satisfactory in all respects and that any and all revisions required by the thesis
examination committee have been made
___________________________________________
Prof Man WONG
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Supervisor
___________________________________________
Prof Libor RUFER
Universiteacute de Grenoble France
Thesis Co-Supervisor
___________________________________________
Prof David COOK
Department of Economics HKUST Hong Kong
Thesis Examination Committee Member (Chairman)
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
A Thesis Submitted to
The Hong Kong University of Science and Technology
in Partial Fulfillment of the Requirements for
the Degree of Doctor of Philosophy
in the Department of Electronic and Computer Engineering
and
Universiteacute de Grenoble
in Partial Fulfillment of the Requirements for
the Degree of Docteur de lrsquo Universiteacute de Grenoble
in the Ecole Doctorale Electronique Electrotechnique Automatique amp Traitement du Signal
February 2013 Hong Kong
iii
Authorization
I hereby declare that I am the sole author of the thesis
I authorize the Hong Kong University of Science and Technology and Universiteacute de
Grenoble to lend this thesis to other institutions or individuals for the purpose of scholarly
research
I further authorize the Hong Kong University of Science and Technology and Universiteacute
de Grenoble to reproduce the thesis by photocopying or by other means in total or in part at
the request of other institutions or individuals for the purpose of scholarly research
___________________________________________
ZHOU Zhijian
February 2013
iv
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
This is to certify that I have examined the above PhD thesis and have found that it is
complete and satisfactory in all respects and that any and all revisions required by the thesis
examination committee have been made
___________________________________________
Prof Man WONG
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Supervisor
___________________________________________
Prof Libor RUFER
Universiteacute de Grenoble France
Thesis Co-Supervisor
___________________________________________
Prof David COOK
Department of Economics HKUST Hong Kong
Thesis Examination Committee Member (Chairman)
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
iii
Authorization
I hereby declare that I am the sole author of the thesis
I authorize the Hong Kong University of Science and Technology and Universiteacute de
Grenoble to lend this thesis to other institutions or individuals for the purpose of scholarly
research
I further authorize the Hong Kong University of Science and Technology and Universiteacute
de Grenoble to reproduce the thesis by photocopying or by other means in total or in part at
the request of other institutions or individuals for the purpose of scholarly research
___________________________________________
ZHOU Zhijian
February 2013
iv
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
This is to certify that I have examined the above PhD thesis and have found that it is
complete and satisfactory in all respects and that any and all revisions required by the thesis
examination committee have been made
___________________________________________
Prof Man WONG
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Supervisor
___________________________________________
Prof Libor RUFER
Universiteacute de Grenoble France
Thesis Co-Supervisor
___________________________________________
Prof David COOK
Department of Economics HKUST Hong Kong
Thesis Examination Committee Member (Chairman)
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
iv
High Frequency MEMS Sensor for Aero-acoustic
Measurements
By
ZHOU Zhijian
This is to certify that I have examined the above PhD thesis and have found that it is
complete and satisfactory in all respects and that any and all revisions required by the thesis
examination committee have been made
___________________________________________
Prof Man WONG
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Supervisor
___________________________________________
Prof Libor RUFER
Universiteacute de Grenoble France
Thesis Co-Supervisor
___________________________________________
Prof David COOK
Department of Economics HKUST Hong Kong
Thesis Examination Committee Member (Chairman)
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
v
___________________________________________
Prof Skandar BASROUR
Universiteacute de Grenoble Grenoble France
Thesis Examination Committee Member
___________________________________________
Prof Wenjing YE
Department of Mechanical Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Levent YOBAS
Department of Electronic and Computer Engineering HKUST Hong Kong
Thesis Examination Committee Member
___________________________________________
Prof Ross MURCH
Department of Electronic and Computer Engineering HKUST Hong Kong
Department Head
Department of Electronic and Computer Engineering
The Hong Kong University of Science and Technology
February 2013
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
vi
Acknowledgments
I would like to give my deepest appreciation first and foremost to Professor Man WONG and
Professor Libor RUFER my supervisors for their constant encouragement guidance and
support though my PhD study at HKUST and Universiteacute de Grenoble Without their
consistent and illuminating instructions this thesis could not have reached its present form
Also I want to thank Professor David COOK for agreeing to chair my thesis examination and
Professor Skandar BASROUR Dr Philippe BLANC-BENON Professor Philippe
COMBETTE Professor Wenjing YE and Professor Levent YOBAS for agreeing to serve as
members of my thesis examination committee
I would like to thank Dr Seacutebastien OLLIVIER Dr Edouard SALZE and Dr Petr
YULDASHEV who are from Laboratoire de Meacutecanique des Fluides et dAcoustique (LMFA
Ecole Centrale de Lyon) and Dr Olivier LESAINT who is from Grenoble Geacutenie Electrique
(G2E lab) the group of Professor Pascal NOUET who is from Laboratoire dInformatique de
Robotique et de Microeacutelectronique de Montpellier (LIRMM lUniversiteacute Montpellier 2) and
Dr Didace EKEOM who is from the Microsonics company (httpwwwmicrosonicsfr) for
their help in guiding the microphone dynamic calibration experiment offering the first
prototype of the amplification card and teaching the ANSYS simulation software under the
project Microphone de Mesure Large Bande en Silicium pour lAcoustique en Hautes
Freacutequences (SIMMIC) which is financially supported by French National Research Agency
(ANR) Program BLANC 2010 SIMI 9
I have appreciated the help of the staffs from the nanoelectronics fabrication facility (NFF)
and materials characterization and preparation facility (MCPF) of HKUST and the technicians
from the Department of Electronic and Computer Engineering and the Department of
Mechanical Engineering of HKUST Also I have appreciated the help of the engineers from
the campus dinnovation pour les micro et nanotechnologies (MINATEC)
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7
vii
Through my PhD study period much assistance has been given by my colleagues and friends
at HKUST I appreciate their kindly help and support and would like to thank them all
especially Ruiqing ZHU Zhi YE Thomas CHOW Dongli ZHANG Parco WONG Zhaojun
LIU Shuyun ZHAO He LI Fan ZENG and Lei LU
During my periods of stay in Grenoble many friends helped me to quickly settle in and
integrate into the French culture I would like to thank them all especially Hai YU Wenbin
YANG Ke HUANG Yi GANG Richun FEI Nan YU Zuheng MING Haiyang DING
Weiyuan NI Hao GONG Zhongyang LI Bo WU Josue ESTEVES Yoan CIVET Maxime
DEFOSSEUX Matthieu CUEFF and Mikael COLIN
Last but not least I devote my deepest gratitude to my parents for their immeasurable support
over the years
viii
To my family
ix
Table of Contents
High Frequency MEMS Sensor for Aero-acoustic Measurements ii
Authorizationiii
Acknowledgments vi
Table of Contents ix
List of Figures xii
List of Tables xvii
Abstract xviii
Reacutesumeacute xx
Publications xxi
Chapter 1 Introduction 1
11 Introduction of the Aero-Acoustic Microphone 1
111 Definition of Aero-Acoustics and Research Motivation 1
[25] C D Lien M A Nicolet and S S Lau Low Temperature Formation of Nisi2 from
Evaporated Silicon in physica status solidi (a) vol 81 WILEY-VCH Verlag pp 123-128
1984
[26] J Zhonghe K Moulding H S Kwok and M Wong The effects of extended heat
treatment on Ni induced lateral crystallization of amorphous silicon thin films Electron
Devices IEEE Transactions on vol 46 pp 78-82 1999
[27] C Hayzelden and J L Batstone Silicide formation and silicide-mediated crystallization
of nickel-implanted amorphous silicon thin films Journal of Applied Physics vol 73 pp
8279-8289 June 15 1993
[28] X F Duan Microfabrication Using Bulk Wet Etching with TMAH MSc Thesis
Department of Physics McGill University 2005
[29] I Virginia Semiconductor Wet-Chemical Etching and Cleaning of Silicon 2003
[30] A E Morgan E K Broadbent K N Ritz D K Sadana and B J Burrow Interactions
of thin Ti films with Si SiO2 Si3N4 and SiOxNy under rapid thermal annealing
Journal of Applied Physics vol 64 pp 344-353 July 1 1988
[31] R W Mann L A Clevenger P D Agnello and F R White Silicides and local
interconnections for high-performance VLSI applications in IBM Journal of Research
and Development vol 39 pp 403-417 1995
[32] L J Chen and E Institution of Electrical Silicide technology for integrated circuits
Institution of Electrical Engineers 2004
[33] Z Yu P Nikkel S Hathcock Z Lu D M Shaw M E Anderson and G J Collins
Measurement and control of a residual oxide layer on TiSi2 films employed in ohmic
contact structures Semiconductor Manufacturing IEEE Transactions on vol 9 pp
329-334 1996
[34] G Yan P C H Chan I M Hsing R K Sharma J K O Sin and Y Wang An improved
76
TMAH Si-etching solution without attacking exposed aluminum Sensors and Actuators
A Physical vol 89 pp 135-141 2001
77
Chapter 4 Testing of the MEMS Sensor
This chapter is divided into four sections The first section presents the testing of key
fabrication process properties including the piezoresistor sheet resistance measurement and
metal to piezoresistor contact resistance measurement The second section presents the static
responses of the microphone samples measured by the nano-indentation technique In the
third section the dynamic calibration method using spark generated shockwave is
demonstrated to measure the frequency response of the wide-band high frequency
microphone And finally the sensor array application as a sound source localizer is presented
41 Sheet Resistance and Contact Resistance
The sheet resistance of the doped MILC poly-Si material was measured by a Greek cross
structure as shown in Figure 41 (also marked within the blue dashed line in Figure 22)
During the test a current IAB was passed through pad A and B and the potential difference
VCD between pad C and D was measured The sheet resistance Rs was calculated using
Equations 41 and 42 shown below
Figure 41 Layout of the Greek cross structure
AB
CD
I
VR (41)
2ln
RRs
(42)
A
B
C
D
78
For the sample fabricated using the surface micromachining technique the measured average
sheet resistances of the sensing area and the connecting area were 4114 ohmsquare (Ω)
and 247Ω respectively For the sample fabricated using the bulk micromachining
technique the measured average sheet resistance of the sensing area was 4464Ω Because
the sensing resistors were fabricated using the same MILC technique with the same impurity
doping and activation conditions for both the surface and bulk micromachining techniques
their sheet resistances are almost the same
The Kelvin structure shown in Figure 42 (also marked within the purple dashed line in Figure
22) was used to measure contact resistance Rc of the metallization system to the doped MILC
poly-Si material During the test a current IAC was passed through pad A and C and the
potential difference VBD between pad B and D was measured The contact resistance was
calculated by Equation 43 and the specific contact resistivity ρc was calculated by Equation
44 where A is the contact area
Figure 42 Layout of the Kelvin structure
AC
BDc I
VR (43)
ARcc (44)
A
B
C
D
79
For the CrAu to MILC poly-Si contact system the measured average contact resistance was
466Ω and the specific contact resistivity was 291μΩmiddotcm2 (with a contact area of 625μm2)
and for the AlSi to MILC poly-Si contact system the measured average contact resistance
was 58Ω and the specific contact resistivity was 232μΩmiddotcm2 (with a contact area of 4μm2)
From this comparison we can see that with the help of the self-aligned titanium silicide layer
the specific contact resistivity of the CrAu to MILC poly-Si system is only a little bit larger
than that of the traditional AlSi to MILC poly-Si system
80
42 Static Point-load Response
The static measurement setup is shown in Figure 43 The fabricated chip was wire-bonded
onto a PCB The latter was then glued to a metallic holder and fixed on a vibration-free stage
A computer-controlled tribo-indentor was used to apply a point-load through a probe with a
conical tip having radius of 25microm (Figure 44) at the center of the sensing diaphragm A
Wheatstone bridge (Figure 45) consisting of two sensing and two reference resistors
respectively on- and off- the diaphragm was used to measure the static force response of the
diaphragm With a DC input bias the output voltage was measured and recorded using an HP
4155 Semiconductor Parameter Analyzer For a 115x115microm2 square diaphragm which was
fabricated using the surface micromachining technique with a DC bias of 2V a static
response of ~04microVVPa was measured (Figure 46) And for a 210x210microm2 square
diaphragm which was fabricated using the bulk micromachining technique with a DC bias of
3V a static response of ~028microVVPa was measured (Figure 47)
Figure 43 Static measurement setup
Figure 44 Cross-sectional view of the probe applying the point-load
PC controller
Triboindentor
(Hysitron)
Sample
Stage
r = 25μm
81
Figure 45 Wheatstone bridge configuration
Figure 46 Typical measurement result with a diaphragm length of 115μm and thickness of 05μm (fabricated using the surface micromachining technique)
Vout
~04microVVPa
Reference resistor
Sensing resistor
Sensing resistor
Reference resistor
82
Figure 47 Typical measurement result with a diaphragm length of 210μm and thickness of 05μm (fabricated using the bulk micromachining technique)
Figure 46 shows that for the surface micromachined device the voltage output is linear at
least to 80μN which is equivalent to 44kPa (equivalent pressure is calculated by dividing the
point force by diaphragm area plus the supporting beamsrsquo areas) And Figure 47 shows that
for the bulk micromachined device the voltage output is linear at least to 160μN which is
equivalent to 36kPa (equivalent pressure is calculated by dividing the point force by
diaphragm area)
Figure 48 and Figure 49 present the applied point-load versus diaphragm center
displacement and corresponding equivalent pressure load versus diaphragm center
displacement relationships respectively The extrapolated mechanical sensitivity in the unit of
nmPa is 032 and 029 for the surface micromachined diaphragm and the bulk
micromachined diaphragm respectively The ratio of the mechanical sensitivity is
032nmPadivide029nmPa =11 while the ratio of the measured static electrical sensitivity is
04microVVPadivide028microVVPa =143 This means that compared to the fully clamped diaphragm
(bulk micromachining technique) the beam supported diaphragm (surface micromachining
technique) has a more efficient mechanical to electrical conversion With the same
displacement the beam supported diaphragm generates more stress at the piezoresistor
~028microVVPa
83
location and leads to a higher electrical voltage output
Figure 48 Point-load vs displacement relationships of sensors fabricated using two different micromachining techniques
Figure 49 Equivalent pressure vs displacement relationships of sensors fabricated using two
different micromachining techniques
84
43 Dynamic Calibration
431 Review of Microphone Calibration Methods
To calibrate a microphone there are many methods with different names However from a
methodology point of view they can be classified into just two categories the primary
method and the secondary method Techniques that are described for calibrating a microphone
except the techniques that require a calibrated standard microphone are considered to be
primary methods A primary method requires basic measurements of voltage current
electrical and acoustical impedance length mass (or density) and time (frequency) In
practice handbook values of density sound speed elasticity and so forth are used rather than
directly measured values of these parameters The secondary methods are those in which a
microphone that has been calibrated by a primary method is used as a reference standard
Secondary methods for calibrating microphones require fewer measurements and provide
fewer sources of error than do primary methods Therefore they are more generally used for
routine calibrations although the accuracy of secondary calibrations can never be better than
the accuracy of the primary calibration of the reference standard if only one standard is used
Accuracy and reliability can be increased by averaging the results of measurements with two
or three standards [1]
4311 Reciprocity Method
The reciprocity method is the mostly used primary method to calibrate microphones The
reciprocity principle as applied to electroacoustics was introduced by Schottky [2] in 1926
and Ballantine [3] in 1929 MacLean [4] and Cook [5] first used it for calibration purposes in
1940 and 1941 The reciprocity theorem is not only valid for the condenser microphone itself
but also for the combined electrical mechanical and acoustical network which is made up of a
transmitter and a receiver microphone coupled to each other via an acoustic impedance This
makes reciprocity calibration possible [6]
85
The reciprocity method requires the to-be-calibrated microphone to be reciprocal that is the
ratio of its receiving sensitivity M to its transmitting response S must be equal to a constant J
called the reciprocity parameter This parameter depends on the acoustic medium the
frequency and the boundary conditions but is independent of the type or construction details
of the microphone To be reciprocal a microphone must be linear passive and reversible
However not all linear passive and reversible microphones are reciprocal Conventional
microphones such as piezoelectric piezoceramic magnetostrictive moving-coil condenser
etc are reciprocal at nominal signal levels [1]
Three-transducer spherical-wave reciprocity is the most commonly used reciprocity method to
calibrate a microphone During calibration the microphones are coupled together by the air
(gas) enclosed in a cavity One microphone operates as a transmitter and emits sound into the
cavity which is detected by the receiver microphone The dimensions of the cavity and the
acoustic impedance of the microphones must be known while the properties (pressure
temperature and composition) of the gas (air) in the coupler must be controlled or monitored
in connection with the measurement These parameters are used for the succeeding
calculations of Acoustic Transfer Impedance and microphone sensitivity Three microphones
(A B and C) are used (Figure 410) They are pair-wise (AB BC and CA) coupled together
For each pair the receiver output voltage and the transmitter input current are measured and
their ratio which is called the Electrical Transfer Impedance is calculated After having
determined the electrical impedance and calculated the acoustic transfer impedance for each
microphone combination the sensitivities of all three microphones may be calculated by
solving the equations below [6]
e ABp A p B
a AB
ZM M
Z (45)
e BCp B p C
a BC
ZM M
Z (46)
e CAp C p A
a CA
ZM M
Z (47)
86
where AB
e ABAB
uZ
i
BCe BC
BC
uZ
i
CAe CA
CA
uZ
i
(MpA MpB MpC pressure sensitivities of microphone A B and C
ZaAB ZaBC ZaCA acoustic transfer impedances of coupler with microphones AB BC and CA
ZeAB ZeBC ZeCA electrical transfer impedances of coupler with microphones AB BC and
CA)
Figure 410 Principle of Pressure Reciprocity Calibration The three microphones (A B and C) are coupled two at a time together by the air (or gas) enclosed in a cavity while the three
ratios of output voltage and input current are measured Each ratio equals the Electrical Transfer Impedance valid for the respective pair of microphones
4312 Substitution Method
The substitution method (also called a comparison calibration method) is a simple secondary
calibration technique When properly made it is reliable and accurate This method consists
of subjecting the to-be-calibrated microphone and a calibrated reference or standard
microphone to the same pressure field and then comparing the electrical output voltages of the
two microphones [6] Theoretically the characteristics of the pressure field generator are
irrelevant It is necessary only that it produces sound of the desired frequency and of a
sufficiently high signal level
iAB
B
A iBC
C
B iCA
A
C
Receivers
Coupler
Transmitters
uAB uBC uCA
87
The standard microphone is immersed in the sound field It must be far enough from the
pressure source that it intercepts a segment of the spherical wave small enough (or having a
radius of curvature large enough) that the segment is indistinguishable from a plane wave
Any nearby housing for preamplifiers or other components must be included in the
dimensions of the microphone because the presence of such housing may affect the
sensitivity
Unless the standard microphone is omni-directional it must be oriented so that its acoustic
axis points toward the pressure source The open-circuit output voltage Vs of the standard
microphone in such a position and orientation is measured The standard microphone then is
replaced by the unknown microphone and the open-circuit output voltage Vx of the unknown
is measured If the free-field voltage sensitivity of the standard is Ms then the sensitivity of
the unknown Mx is found from the following
xx s
s
VM M
V (48)
A variation of the substitution method is the practice of simultaneously immersing both the
standard and the unknown microphone in the medium and in the same sound field (also
named the simultaneous method) Since the two microphones cannot be in the same position
this technique requires some assurance that the sound pressure at the two locations is the same
or has some known relationship If the microphones are placed close together the presence of
one may influence the sound pressure at the position of the other and if the microphones are
placed far apart reflections from boundaries and the directivity of the pressure source may
produce unequal pressure at the two locations If the boundary and medium conditions are
stable the relationship between the sound pressures at the two locations can be measured The
disadvantages of this variation usually outweigh the advantages and the method is not used
very much
88
4313 Pulse Calibration Method
The reciprocity and substitution methods are well established to calibrate microphones in the
audio frequency range (20Hz ~ 20kHz) However they are difficult to apply in the wide-band
high frequency microphone calibration area As we described in the previous section the
microphone produced in this thesis is original and unique which means no comparable
microphone exists on the market Therefore no commercial standard microphone can be used
as the reference in the substitution calibration method and this microphone can not be
calibrated by the secondary method Reciprocity is a primary method However that the
microphone be reciprocal is a prerequisite and the piezoresistive type aero-acoustic
microphone does not meet this requirement
The most difficult part of the primary calibration process is to know the exact pressure (force)
applied to the microphone diaphragm In the audio frequency range this is achieved by using
a piston-phone which provides a constant and known volume velocity to a microphone and
in the lower ultrasonic frequency band (up to 100kHz) an electrostatic actuator (EA) is
normally used to apply a known force to the microphone The EA produces an electrostatic
force which simulates sound pressure acting on the microphone diaphragm In comparison
with sound based methods the actuator method has a great advantage in that it provides a
simpler means of producing a well-defined calibration pressure over a wide frequency range
without the special facilities of an acoustics laboratory However the EA method requires an
accessible conductive diaphragm [7] which is not compatible with some kinds of
microphones including the piezoresistive type
There is no single tone wide-band pressure source (gt 100kHz) on the market and the simple
reason is that no wide-band high frequency microphone in this range could be used to
calibrate the source Much work has been done in the calibration of acoustic emission (AE)
devices in the ultrasonic range These use a pulse or step force such as the Hsu-Nielsen
method (also named as pencil lead breaking method) [8] or glass capillary breaking method
[9] as a source Figure 411 presents three kinds of pulse signal and their fast Fourier
89
transform The basic idea of these methods is that the smaller the pulse duration is the wider
the flat band pressure that can be generated from the system
Figure 411 Pulse signals and their corresponding spectra
Hsu-Nielsen and glass capillary breaking methods could not be directly used for the
wide-band high frequency microphone calibration since they generate a pulse signal in the
form of displacement which is only suitable for an AE sensor Considering the microphone
calibration a pulse signal in the pressure form should be generated and more specifically the
pressure pulse duration should be in the micro-second range which makes the frequency
bandwidth ~1MHz and the pressure level should be adjustable for a large dynamic range
which matches the microphone specifications Table 41[7] summarizes the methods to
calibrate a microphone Until now the pulse calibration method has been the most suitable for
a wide-band high frequency microphone
Pulse calibration method
Requires pulse duration in micro-second range
Pulse
Time [s]
Amplitude
Single-side frequency spectrum
Frequency [Hz]
90
Table 41 Summary of different microphone calibration methods
Method Bandwidth Limitations
Reciprocity Low frequency Microphone to be reciprocal
Substitution Low frequency Need calibrated reference
Piston-phone Low frequency Limited sound pressure level
EA High frequency Need conductive diaphragm
Pulse High frequency Not mature technique
432 The Origin Characterization and Reconstruction Method of N Type
Acoustic Pulse Signals
Figure 412 presents an ideal N type acoustic pulse signal (N-wave) in 10μs duration and its
corresponding frequency spectrum Even though the frequency spectrum is not flat it still
could be used as a pulse source to calibrate microphones The work has been verified by
Averiyanov [10]
Figure 412 An ideal N-wave in 10 μs duration and its corresponding frequency spectrum
91
4321 The Origin and Characterization of the N-wave
The origin of the discovery of the N-wave is from the study of small firearm bullets (Figure
413) but it has been found that the same mathematical expressions will describe the
characteristics of the N-wave as a good approximation for supersonic projectiles of any sizes
and shapes [11]
Figure 413 N-wave near projectile (a) Cone-cylinder (b) Sphere
Although the N-wave starts as a wave with considerably rounded contours as illustrated
schematically in Figure 414(a) it rapidly changes into an N shape wave such as that shown in
92
Figure 414(c) This is due to the fact that the particles of the medium in the compressed
portions of the wave are traveling noticeably faster than normal sound velocity while the
particles in the rarefaction phase are traveling at slower velocities Consequently the high
positive amplitudes arrive early at a given point and the high negative amplitudes arrive late
Thus the wave steepens to have a sudden sharp rise gradually diminishes to a point below
the ambient pressure and then suddenly recovers to ambient pressure at the end
(a) Start (b) Intermediate (c) Final
Figure 414 N-wave generation process
To study and characterize the N-wave it is good to use a full scale model which means that
when the generated N-wave is characterized the original source is used This is still possible
or affordable for the N-wave source study which will not cost too much However when it is
used as an acoustic source for microphone calibration the cost will directly limit the number
of trials and the results will also be affected by environmental factors such as the temperature
humidity background noise etc To get a more cost effective and repeatable N-wave
researchers have tried to build an artificial N-wave source for which the generation conditions
can be easily controlled in a laboratory
Many techniques have bean investigated to generate the N-wave under laboratory scale
conditions The simplest way to generate the N-wave is from the bursting of a balloon [12]
When an initial spherical uniform static-pressure distribution is released the acoustic
disturbance that results has the N shape which is predicted from the linear acoustic-wave
equation with the appropriate boundary conditions Generally two methods can be used to
burst the balloon The first method is to fill the balloon with air until it ruptures spontaneously
and the second one is to fill the balloon with air seal it off just before the breaking point and
puncture it with a pin or any sharp object Experiments show that the spontaneous rupture
93
tears the balloon into many small shreds indicating a more complete disintegration of the skin
Thus this method results in a closer approximation of a pressure distribution which is
released at all points
A similar method but with better controlled equipment is the shock tube (Figure 415) which
can be used to generate the N-wave under laboratory scale conditions also [13] It consists
basically of a rigid tube divided into two sections These sections are separated by a gas-tight
diaphragm which is mounted normally to the axis Initially a significant pressure difference
exists between the two sections The high pressure section is called the compression chamber
while the low pressure section is known as the expansion chamber When the diaphragm is
ruptured the pressure begins to equalize in the form of a shock wave (N-wave) moving into
the expansion chamber and a rarefaction wave moving into the compression chamber
Figure 415 Schematic of the shock tube
Other methods such as using a laser as a focused electromagnetic energy source to burn the
target and generate the N-wave have also been reported [14-16] However the most
commonly used method is generation from a high voltage electrical spark This method is a
robust way to generate an intense acoustic pulse that acts independently of the acoustic
matching between the emitter and medium It is far less sensitive to any contamination In
addition the directivity pattern is essentially omni-directional in the equatorial plane and the
acoustic characteristics have proven to be repeatable for successive sparks Studies on the
acoustic wave that occurs after a spark discharge in air have been performed [17 18] and this
method is even used to act as an ultrasonic generator in the flow measurement situation [19]
A simple spark discharge circuit is shown in Figure 416 [20] A high voltage power supply
Compression chamber Expansion chamber
Diaphragm
Pressurization valve Release valve
94
(~14kV) charges a storage capacitor (1nF) through a current limiting resistor (50MΩ) and the
discharge of the capacitor occurs through the spark gap (~13cm) which may reach one ohm
of resistance or less during discharge The process of electrical breakdown may be outlined as
follows When the voltage across the gap reaches a sufficiently high potential (breakdown
voltage) causing ionization in the air around the gap a very narrow cylindrical region
between the gap becomes a good conductor The energy stored in the circuit surges through
this region often raising the temperature to several thousand degrees Kelvin This results in
the rapid expansion of the spark channel forming a cylindrical shock ahead of it The initial
shock usually pulls away from the spark channel within 1 micro-second and the shock front
is first observed to be ellipsoidal with its major axis along the axis of the spark Within 10
micro-seconds however it assumes a nearly perfect spherical shape
Figure 416 High voltage capacitor discharge scheme
Figure 417 shows an ideal N-wave generated by the electrical spark discharge which is
characterized by two parameters the half duration T and the overpressure Ps The intensity of
the spark is controlled by the electrical energy stored in the capacitor
20
1
2E CV (49)
where E0 is the stored electrical energy C is the capacitor for energy storage and V is the
charging voltage By simplifying the spark source to appear as a point source producing a
~14kV
1nF
50MΩ
Spark gap ~13cm
95
spherical omni-directional wave at normal room temperature Wyber [18] theoretically
estimates the electrical to acoustical energy transform efficiency to be ~007 as shown in
Equation (410)
0007AE E (410)
where EA is the generated acoustical energy from the electrical spark discharge in the unit of
joule
Plooster [21] characterizes the relationship between the overpressure and the released energy
in Equation (411
2
2
( 1)u
s
EP
b r
(411)
where Eu is the energy released per unit length of the source γ is the air specific heat ratio
which is equal to 14 b is a parameter which is only dependent upon γ and is found to be 394
r is the distance between the location of the calculated overpressure and the source and δ is
unity under the strong shock solution
The half duration T is proportional to the spark gap distance To summarize the acoustic
overpressure generated by the electrical spark discharge is proportional to the released energy
The larger spark gap needs higher voltage to break down the air which leads to larger
released energy and in turn a higher acoustic overpressure But on the other hand the larger
spark gap will also lead to a larger half duration of the N-wave which will limit the frequency
information A typical spark with ~11us half duration and 23kPa overpressure at 10cm
propagation distance is recorded by Wright [17]
96
Figure 417 Schematic of an ideal N-wave
4322 N-wave Reconstruction Method
To accurately calibrate a microphone it is important to know the exact shape of the N-wave
generated in our laboratory conditions Figure 418 presents a real N-wave and the shape of
this real N-wave is decided by three parameters the half duration T the overpressure Ps and
the rise time t (defined as the time interval from 10Ps to 90Ps)
The rise time t of the N-wave is measured by focused shadowgraphy By using the
shadowgraphy technique the distribution of light intensity in space is photographed and then
analyzed The pattern of the light intensity is formed due to the light refraction in
non-homogeneities of the refraction index caused by variations of medium density Shadow
images called shadowgrams are captured by a camera at some distance from the shock wave
by changing the position of the lens focal plane
The setup designed for this optical measurement is shown in Figure 419 [22] It is composed
of a 15kV high voltage spark source which is used to generated an acoustic N-wave a BampK
wideband microphone (type 4137 cut-off frequency ~200kHz) which is used to determine
the N-wave amplitude and duration and deduce the theoretical rise time using a Generalized
Time
Pressure
Ps
97
Burgers equation and optical equipment including a flash-lamp light filter lens and a digital
CCD camera These pieces of optical equipment were mounted on a rail and aligned coaxially
The flash-lamp generated short duration (20ns) light flashes that allowed a good resolution of
the front shock shadow The focusing lens was used to collimate the flash light in order to
have a parallel light beam The dimension of the CCD camera was 1600 pixels along the
horizontal coordinate and 1186 pixels along the vertical coordinate The lens was used to
focus the camera at a given observation plane perpendicular to the optical axis Compared to
the rise time deduced from the microphone measurement the optical measurement result
matches better with the theoretical estimation (Figure 420) [22] which verifies that the rise
time result is limited by the frequency bandwidth of the microphone used
Figure 418 Real N-wave shape
T
t
Ps
98
Figure 419 Shadowgraph experiment setup (1 spark source 2 microphone in a baffle 3 nanolight flash lamp 4 focusing lens 5 camera 6 lens)
Figure 420 Comparison between the optically measured rise time and the predicted rise time by using the acoustic wave propagation at different distances from the spark source
The half duration T of the N-wave normally around 20μs which equivalents to 25kHz in
frequency spectrum can be directly measured by a BampK microphone type 4138 with a
bandwidth of 140kHz
99
To know the overpressure Ps0 at distance r0 from the spark source at first the half duration T0
at distance r0 is measured Then by varying the distance r a series of N-wave half duration
values T at corresponding distance r are recorded For a spherical N-wave weak shock theory
gives the following evolution law for the half duration [23]
000 ln1)(
r
rTrT (412)
00
000 2
)1(
TcP
Pr
atm
s
(413)
where γ = 14 is the ratio of the specific heat for gas Patm is the atmospheric pressure and c0 is
the sound speed From Equation (412) the coefficient σ0 shows the dependence of half
duration T to the initial overpressure at distance r = r0 As we have already recorded a series
of half duration T at different distances r the parameter (TT0)2-1 is plotted as a function of
ln(rr0) Then the slope of the linear fitted line is the coefficient σ0 Once the coefficient σ0 is
obtained the overpressure Ps0 can be calculated by Equation (414)
0
0000 )1(
2
r
TcPP atm
s
(414)
433 Spark-induced Acoustic Response
As we found from the static nano-indentation measurement the sensitivity of the sample is
very low So an amplification card was connected to the sensor output to boost the signal and
make it large enough for the oscilloscope to capture Figure 421 shows the schematic of the
amplification card connecting to the sensor The card is composed of a two-stage
configuration with two identical instrumentation amplifiers (INA103) The first stage is a
pre-amplifier directly connected to the sensor output with a gain of 10 A high pass filter with
-3dB cut-off frequency at 1kHz is inserted in between the first stage and the second stage (C =
10nF R = 15kΩ) This high pass filter blocks the possibly amplified DC off-set signal
100
originally from the sensor to prevent voltage saturation of the second stage which has a large
gain of 100 The frequency response of the amplification card is shown in Figure 422 With a
real gain of 58dB the -3dB cut-off frequency is 600kHz
Figure 421 Schematic of the amplifier
Figure 422 Frequency response of the amplification card
The dynamic calibration setup is shown in Figure 423 The spark discharging circuit is
configured the same as Figure 416 The microphone sample is glued to a PCB and wire
bonded The PCB is then put into a baffle which is used to eliminate the acoustic reflection
Sensor Pre-amplification Filter Amplifier
101
effect due to the PCB edge The baffle is specially designed so that it allows the PCB to be
surface mounted into it (Figure 424) The gap between the PCB and the surrounding baffle is
covered by Scotch tape
Figure 423 Spark calibration test setup
Figure 424 Baffle design
The amplification card was put into an aluminum shielding box which prevented the strong
electromagnetic interference generated by the electrical discharge The to-be-calibrated
microphone sample was connected to the amplification card through a small hole in the
shielding box front surface Finally the shielding box was placed on top of a stage which
could move along the guided rail and be controlled through LabVIEW software
Baffle PCB
Microphone sample Scotch tape
Spark generator
Shielding box
Microphone sample
with baffle
102
4331 Surface Micromachined Devices
After discovering the exact N-wave shape at distance r0 away from the spark source our
to-be-calibrated samples were placed at the same distance A typical measured N-wave signal
using surface micromachining devices is shown in Figure 425 From the figure we can
clearly find two consecutive oscillation signals The first oscillation corresponds to the sharp
rise of the front shock of the N-wave and the second oscillation corresponds to the sharp rise
of the rear shock of the N-wave However the low frequency information of the N-wave
corresponding to the slope from the front shock to rear shock cannot be seen in the measured
curve This also verifies the low frequency information loss due to the acoustic short path
effect which is predicted in the finite element modeling At the same time we find that due to
the fact that this device is only sensitive to the high frequency signal which is related to the
sharp upward rise step in the signal time domain both the first and second measured
oscillations start with an upward curve The single-sided spectra of the measured signals from
the microphone and from the optical method are obtained by applying fast Fourier transform
(FFT) to the time domain signals (Figure 426) The frequency response (electrical sensitivity
in the unit of VPa) is defined by Equation 415 When using decibel (dB) in the logarithmic
unit (referring to 1VPa) Equation 415 is changed to Equation 416 and the frequency
response can be calculated by directly subtracting the green curve in Figure 426 from the
The frequency response of the calibrated microphone is shown in Figure 427 which is also
compared with FEA result The resonant peak is about 400kHz which is the same as the
103
prediction of the FEA result The flat band is very narrow roughly from 100kHz to 200kHz
and below 100kHz the frequency response is quickly decreased The dynamic sensitivity
within the flat band is 0033microVVPa which is much lower than the static value (04microVVPa)
This phenomenon could also be explained by the acoustic short path effect (Figure 428)
Using the N-wave reconstruction method we can accurately find the incident pressure P0 to
the sensing diaphragm But the real pressure difference ∆p on the sensing diaphragm is equal
to P0 ndash Ps (Ps is the leaked pressure into the air cavity through the release holesslots) which is
difficult to predict
Figure 425 Typical spark measurement result of a microphone sample fabricated using the surface micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 426 FFT single-sided amplitude spectra of the measured signals from a surface micromachined microphone and from the optical method
fr = 400kHz
104
Figure 427 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with FEA result
Figure 428 Acoustic short circuit induced leakage pressure Ps
Thermal oxide Amorphous silicon
Low stress nitrideMILC poly-Si
TiSi Metallization
P0
Ps
Incident wave
105
4332 Bulk Micromachined Devices
Figure 429 shows the typical measured N-wave signal using bulk micromachining devices
and Figure 430 presents the corresponding spectra calculated using the FFT algorithm From
Figure 430 we can see that the bulk micromachining devices have a larger resonant
frequency (715kHz) and from Figure 429 we can see that not only the high frequency
information but also the low frequency information can be caught by this device (the slope
from the front shock of the N-wave to the rear shock of the N-wave) Also we can see that
there is an oscillation superimposed on the slope which means that the microphone device is
not sufficiently damped at its resonant frequency
Figure 429 Typical spark measurement result of a microphone sample fabricated using the bulk micromachining technique (3V DC bias with amplification gain 1000 and source to
microphone distance is 10cm)
Figure 430 FFT single-sided amplitude spectra of the measured signals from a bulk micromachined microphone and from the optical method
fr = 715kHz
106
Again using the calculation method mentioned in the previous section the frequency
response of the bulk micromachining devices is shown in Figure 431and is compared with
the lumped-element modeling result The dynamic sensitivity is 1mVPa (with amplification
gain 1000 and 3V DC bias) which means that the real microphone dynamic sensitivity is
about 033microVVPa and is similar to the static calibrated sensitivity (028microVVPa) Also this
microphone has a wide flat bandwidth from 6kHz up to 500kHz However compared to the
lumped-element model the measured resonant frequency is a little smaller This phenomenon
is possibly caused by the LS-SiN material properties variations between different fabrication
batches The material properties used in the lumped-element modeling were measured from
the test batch while the real device was fabricated 6 months later
Figure 431 Frequency response of the calibrated microphone (3V DC bias with amplification gain 1000 averaged signal) compared with lumped-element modeling result
Finally the spark measurement results of these two microphones compared with the optical
measured signal are shown in Figure 432 and the comparison of the frequency responses are
presented in Figure 433
107
Figure 432 Comparison of the spark measurement results of microphones fabricated by two different techniques (spark source to microphone distance is 10cm)
Figure 433 Comparison of the frequency responses of microphones fabricated by two different techniques
108
44 Sensor Array Application as an Acoustic Source Localizer
To calculate an acoustic source in a Cartesian coordinate system (Figure 434) with three
unknown parameters x y and z we need three equations to solve (as shown in Equation 417)
where (x y z) are the acoustic source coordinates (xii=123 yii=123 zii=123) are the three
sensor coordinates and dii=123 are the distances between the acoustic source and each sensor
These distances are calculated using Equation 418 where v is the sound velocity and tii=123
are the acoustic waversquos travelling time from the source to each sensor
Figure 434 Cartesian coordinate system for acoustic source localization
23
23
23
23
22
22
22
22
21
21
21
21
)()()(
)()()(
)()()(
dzzyyxx
dzzyyxx
dzzyyxx
(417)
vtd
vtd
vtd
33
22
11
(418)
y
x
z
(x2y2z2) (x1y1z1)
(x3y3z3)
t2d2 t1 d1
t3 d3
(xyz) Acoustic source
M1 M2
M3
Origin
point
109
Three sensors were placed in one plane to form an array as shown in Figure 434 and Figure
435 The first sensor (M1) has a coordinate of x1 = 25 y1 = 0 and z1 = 0 the second sensor
(M2) has a coordinate of x2 = -25 y2 = 0 and z2 = 0 and the third sensor (M3) has a coordinate
of x3 = 0 y3 = 4 and z3 = 0 all in the unit of centimeter
Figure 435 Sensor array coordinates
The sound velocity v is a key parameter in the coordinate calculation process and it is
sensitive to the environmental parameters such as ambient pressure temperature and
humidity So before location coordinate calculation the sound velocity v should be well
calibrated The acoustic source was fixed at the XY plane (xo yo) with Z coordinate zo = 0 and
one microphone was placed with the same X and Y coordinates (xo yo) while the Z coordinate
zm changed from 10cm to 105cm (Figure 436) The acoustic signal captured by the sensor
was recorded by an oscilloscope The acoustic source was the spark generator as mentioned
in the previous section and the oscilloscope was triggered by the electromagnetic signal from
the spark As the electromagnetic signal travels at a speed of 3times108ms which is much faster
than the speed of sound the sound travelling time was calculated using the delay time
between the oscilloscope trigger point time and the recorded signal arrival time
The sound travelling distance vs travelling time is shown in Figure 437 The velocity is
extrapolated by linearly fitting the measured data and the value is 3442ms From the linear
M1 M2
M3
X
Y
0
110
fitting curve we also find an offset of 21mm when time is equal to zero which could come
from a system setup error
Figure 436 Sound velocity calibration setup
Figure 437 Sound velocity extrapolation
Figure 438 presents the setup for the acoustic source localization application The spark
generator emitted an acoustic wave which was sensed by the sensor array The sensed signals
were captured by an oscilloscope (Tektronix TDS 2024C) and then the captured signals were
transferred to a laptop through a USB cable using the MATLAB Instrument Control Toolbox
which is based on the National Instruments Virtual Instrument Software Architecture
(NI-VISA) standard Then the delay times and the acoustic source coordinates were calculated
by MATLAB software All of these functions were realized by a customized MATLAB
graphic user interface (GUI)
Acoustic source Sensor
0 Z
(xo yo zo = 0) (xo yo zm = 10~105cm)
111
Figure 438 Acoustic source localization setup
During the GUI initialization firstly the sound velocity was required to be input otherwise
the default value of 340ms would be used (Figure 439) After initialization the main window
as shown in Figure 440 popped up The main window consists of three parts the main
figures showing the captured acoustic signals and source locations projected in the XY plane
(marked by the red dashed line in Figure 440) the boxes showing the calculated delay times
of each signal the input sound velocity and the calculated source coordinates (marked by the
pink dashed line in Figure 440) and session log information and functional buttons (marked
by the blue dashed line in Figure 440) The ldquoConnectionrdquo button was used to initialize the
communication between the GUI and the oscilloscope and the ldquoStartrdquo button was used to
initiate the data transfer from the oscilloscope to the MATLAB software and the following
data processing
Figure 439 GUI initialization for sound velocity input
Sensor array
0 Z
Sound source
112
Figure 440 Localization GUI main window
113
During the localization test the spark source was fixed at one position and the sensor array
was moving in the Z direction But the origin of the Z coordinate was always the sensor array
plane as shown in Figure 434 and Figure 441 which was equivalent to the setup in which
the sensor array was fixed at the coordinate origin and the sound source was moving The
reason for this setup arrangement is simply that the high voltage cable connecting the voltage
generator and spark needles is not long enough
Figure 441 Localization test of the Z coordinate system
The spark sound source was preset at the coordinates of (xs = 0cm ys = 4cm) in the XY plane
Because the two spark needles had a gap of 13cm the middle position of the gap was
assumed to be the source position (Figure 442) The distance between the sound source and
the sensor array in the Z coordinate was changing from 10cm to 105cm (the distance was
measured by a ruler) At each position 20 measurements were carried out Using the
measured delay times the calibrated sound velocity and using Equation 417 and Equation
418 the sound source coordinates were calculated and compared with the values which were
pre-measured by a ruler (Figure 443)
Figure 442 Sound source position definition
Sound source Sensor array plane
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
YAssumed source position
114
Figure 443 Coordinates comparisons between the pre-measured values and the calculated values (a) X coordinates (b) Y coordinates and (c) Z coordinates
Figure 443 shows that the pre-measured values and the calculated values of the Z coordinates
matched very well while the X and Y coordinates did not For the X coordinates (Figure
443(a)) the calculated values fluctuated around the pre-measured values This phenomenon
could be explained by the fact that the real spark generation point was not always at the
middle of the two needles the point varied during the experiment and was different from
position to position To verify this assumption a high speed camera is needed to capture the
(c)
(b)
(a)
115
spark images during the whole measurement process for position analysis which is not
applicable at the current stage
For the Y coordinates (Figure 443(b)) the differences between the pre-measured values and
the calculated values linearly increased up to 2cm when the measurement position changed
from 1 to 11 (from Figure 443(c) this position changing means the Z coordinates changed
from 10cm to 105cm) There are three possible reasons that may explain this phenomenon
One reason is that the table surface onto which the measurement setup was placed was not
level the second reason is that the ground surface was not level and the third is the
combination of the previous two effects Table 42 presents the measured distance between the
table surface and ground surface at corresponding measurement positions These results
eliminate the possibility that the table surface was unlevel So the differences between the
pre-measured values and the calculated values of the Y coordinates can be explained by the
ground surface being unlevel as shown in Figure 444 The angle θ between the ground
surface and the level is calculated to be 11deg
Table 42 Distance between table surface and ground surface at different positions
[20] R E Klinkowstein A study of acoustic radiation from an electrical spark discharge in
air MS Thesis Department Mechanical Engineering Massachusetts Institute of
Technology 1974
[21] M N Plooster Shock Waves from Line Sources Numerical Solutions and Experimental
Measurements Physics of Fluids vol 13 pp 2665-2675 November 1970
[22] P Yuldashev M Averiyanov V Khokhlova O Sapozhnikov S Ollivier and P Blanc
Benon Measurement of shock N-waves using optical methods in 10eme Congres
Francais dAcoustique Lyon France 2010
[23] S Ollivier E Salze M Averiyanov P V Yuldashev V Khokhlova and P Blanc-Benon
Calibration method for high frequency microphones in Acoustics 2012 conference
119
Chapter 5 Summary and Future Work
51 Summary
In this thesis at the beginning the definition and the performance specifications of the
wide-band aero-acoustic microphone were introduced This kind of microphone is specifically
used in the acoustic scaled modeling technique with a scaling factor M larger than 20 which
requires the microphone to have a bandwidth of several hundreds of kilo-Hz and a dynamic
range of up to 4kPa Then a comparative study of the current state-of-the-art of capacitive
and piezoresistive microphones especially the study of their scaling properties demonstrated
that a piezoresistive sensing mechanism is more suitable for achieving the wide-band and
large sensitivity requirements
In Chapter Two first the key mechanical properties including residual stress density and
Youngrsquos modulus of LS-SiN which was used to build the sensing diaphragm were discussed
and measured Following this the design considerations due to the use of different
micro-fabrication techniques (surface micromachining technique and bulk micromachining
technique) were discussed and two different mechanical structures were proposed and
modeled by the FEA method at the end of the chapter
Because the piezoresistive material is the same for both micromachining techniques at the
beginning of Chapter Three a review of the material fabrication technique (MILC) was
presented Then detailed fabrication processes of the surface micromachining and bulk
micromachining techniques were illustrated with transitional schematic views of the
microphone cross-sectional areas
In Chapter Four firstly the electrical performances of the piezoresistor such as sheet
resistance and contact resistance were measured Then the static point-load response was
measured using the nano-indentation technique Following this the microphone dynamic
120
calibration methods including the reciprocity method substitution method and pulse
calibration method were reviewed Due to the characteristics of the piezoresistive sensing
mechanism and commercial reference microphone market limitations both the reciprocity and
substitution methods are not suitable for calibrating these newly designed wide-band high
frequency microphones Only pulse calibration which requires a repeatable high acoustic
amplitude and short duration acoustic pulse source is suitable for our calibration process
Then the acoustic pulse source an electrical discharge induced spark generator was
presented and the characterization and reconstruction method of the generated N-wave were
introduced Finally the dynamic calibrated microphone frequency responses were shown and
compared
Comparisons between other already demonstrated piezoresistive type aero-acoustic
microphones and the current work are listed in Table 51 While keeping a small diaphragm
size the microphone in the current work achieves the highest measurable pressure level at
least up to 165dB and has the widest calibrated bandwidth from 6kHz to 500kHz This
microphone has a lower sensitivity The main reason is that the sensing material used in the
current work is MILC poly-Si material which has a lower gauge factor compared to the sc-Si
material used in Arnold and Sheplakrsquos work Another reason is that the piezoresistor geometry
shape in the current work is not optimized especially the piezoresistor thickness To make the
resistance of the piezoresistor smaller which means the electrical-thermal noise is smaller
(Equation 51) the piezoresistor thickness is kept relatively large This makes the maximum
diaphragm bending stress be not at the diaphragm surface where the piezoresistor is located
4th BS K RT 2[ ]V Hz (51)
( BK is the Boltzmann constant R is the resistance and T is the temperature in Kelvin)
121
Table 51 Comparisons of current work and state-of-the-art
Microphone Type Radius
(mm)
Max pressure
(dB)
Sensitivity Bandwidth
(predicted)
Arnold et al
[1]
piezoresistive 05 160 06μVVPa(3V) 10Hz ~ 19kHz
(~100kHz)
Sheplak et al
[2]
piezoresistive 0105 155 22μVVPa(10V) 200Hz ~ 6kHz
(~300kHz)
Current work piezoresistive 0105
(square) 165 028 mVVPa (3V) 6kHz (DC)
~500kHz
122
52 Future Work
Although two wide-band high frequency microphone prototypes were successfully fabricated
and calibrated there are several issues that need to be worked on in the near future Firstly
models of these two microphones are all based on the FEA method This method is useful and
accurate for structure performance verification but the limitation is that it is not suitable to
use for design which means that given specifications a designer needs to conduct many
trials to find the structurersquos shape and dimensions Therefore an analytical model which may
not be accurate but could quickly estimate the performance of different structures is urgently
needed
Secondly for the microphone fabricated using the bulk micromachining technique due to the
large cavity under the sensing diaphragm there is no sufficient damping to critically damp the
resonant peak In the future a new structure with an integrated damper using the squeeze film
damping effect should be explored At the same time as the titanium silicidation technique is
not needed for reducing contact resistance the thickness of the piezoresistor could be
decreased to increase the sensitivity The trade-off between increasing sensitivity and
increasing noise level due to the decreasing of the piezoresistorrsquos thickness should be
optimized
Thirdly in our testing the amplifier is built by discrete components on the PCB and the
sensor and amplifier are connected through wire bonding To depress the noise and increase
the amplification performance the amplifier should be fabricated on one chip and eventually
the sensor and amplifier should be fabricated on one die together
123
53 References
[1] D P Arnold S Gururaj S Bhardwaj T Nishida and M Sheplak A piezoresistive
microphone for aeroacoustic measurements in Proceedings of ASME IMECE 2001
International Mechanical Engineering Congress and Exposition pp 281-288 2001
[2] M Sheplak K S Breuer and Schmidt A wafer-bonded silicon-nitride membrane
microphone with dielectrically-isolated single-crystal silicon piezoresistors in
Technical Digest Solid-State Sensor and Actuator Workshop Transducer Res
Cleveland OH USA pp 23-26 1998
124
Appendix I Co-supervised PhD Program Arrangement
My PhD study was co-supervised by Dr Man WONG Professor of Electronic and Computer
Engineering (ECE) Department at the Hong Kong University of Science and Technology
(HKUST) and Dr Libor RUFER Researcher at Laboratoire Techniques de lrsquoInformatique et
de la Microeacutelectronique pour lrsquoArchitecture des systegravemes integers (TIMA Lab) France Dr
RUFER is also affiliated with Centre national de la recherche scientifique (CNRS France)
Universiteacute Joseph Fourier (UJF) and Grenoble Institute of Technology (Grenoble-INP) In
June 2009 UJF Grenoble-INP and other research institutes merged into Universiteacute de
Grenoble (UG) so I registered both in the HKUST and UG from 2009 to 2013
My research work was financially supported by the French Consulate at Hong Kong and also
funded by Agence Nationale de la Recherche (ANR French National Agency for Research)
through Program BLANC 2010 SIMI 9 for the project SIMMIC The consortium for this
project consisted of three academic laboratories TIMA LIRMM (Laboratoire dInformatique
de Robotique et de Microeacutelectronique de Montpellier lUniversiteacute Montpellier 2) and LMFA
(Laboratoire de Mecanique des Fluides et dAcoustique Ecole Centrale de Lyon) and one
private partner (Microsonics)
For my PhD study generally speaking when I was in Hong Kong research works were
estimating the mechanical vibration of the sensing diaphragm using lumped-element model
and FEA method developing the corresponding sensor fabrication process and preliminary
static response measurement I spent one year in Grenoble from February 2011 to July 2011
and February 2012 to July 2012 When I was in Grenoble research works were sensor
dynamic calibration with the cooperation of LMFA and sensor mechanical-acoustic
interaction modeling with the cooperation of Microsonics
125
Appendix II Extended Reacutesumeacute
Pour les raisons de clarteacute et de faciliteacute de compreacutehension dans cette thegravese le capteur MEMS
agrave haute freacutequence sera eacutegalement deacutenommeacute le microphone MEMS aeacutero-acoustique agrave large
bande Lrsquoaeacutero-acoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit soit
par un mouvement turbulent du fluide soit par les forces aeacuterodynamiques qui interagissent
avec les surfaces Lrsquoaeacutero-acoustique est un secteur en croissance qui attire une attention
contemporaine en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale
Conformeacutement agrave la deacutefinition ci-dessus notre recherche se concentre principalement sur trois
domaines aeacutero-acoustiques Tout dabord des avanceacutees significatives en aeacutero-acoustique sont
neacutecessaires pour reacuteduire le bruit environnemental et le bruit de cabine geacuteneacutereacutes par les avions
subsoniques et pour se preacuteparer agrave leacuteventuelle entreacutee agrave grande eacutechelle des avions
supersoniques dans laviation civile Dautre part dans le domaine des transports terrestres les
efforts sont faits pour reacuteduire le bruit aeacuterodynamique des automobiles et des trains agrave grande
vitesse Enfin si le bruit des veacutehicules lanceacutes dans lrsquoespace nest pas controcircleacute de graves
dommages structurels peuvent ecirctre engendreacutes au veacutehicule et agrave sa charge
Alors que les testsmesures dun objet dans une situation reacuteelle sont possibles leur deacutepense est
trop eacuteleveacutee leur configuration est geacuteneacuteralement compliqueacutee et les reacutesultats sont facilement
corrompus par le bruit ambiant et par les changements de paramegravetres environnementaux tels
que les fluctuations de la tempeacuterature et de lhumiditeacute Par conseacutequent les tests effectueacutes en
laboratoire dans une condition bien controcircleacutee en utilisant les modegraveles de dimension reacuteduite
sont preacutefeacuterables
La plupart des travaux anciens sur les microphones MEMS ont porteacute sur la conception des
microphones acoustiques low-cost pour leurs applications dans la teacuteleacutephonie mobile En
revanche lobjectif de cette thegravese est clairement axeacute sur les applications meacutetrologiques en
acoustique dans lrsquoair et plus particuliegraverement sur les applications acoustiques du modegravele
reacuteduit ougrave les mesures preacutecises des ondes de pression agrave large bande avec une freacutequence de
126
plusieurs centaines de kHz et les niveaux de pression allant jusquagrave 4kPa sont essentielles
Afin de couvrir une large gamme de freacutequences les transducteurs eacutelectro-acoustiques pour la
geacuteneacuteration et la deacutetection de signal acoustique dans lair utilisent traditionnellement les
eacuteleacutements pieacutezoeacutelectriques Les transducteurs pieacutezoeacutelectriques classiques en volume vibrant en
mode deacutepaisseur ou de flexion ont eacuteteacute largement utiliseacutes pour les deacutetecteurs de preacutesence Lun
des inconveacutenients de ces systegravemes est la neacutecessiteacute dutiliser les couches dadaptation sur la
surface active du transducteur ce qui minimise la diffeacuterence principale entre lrsquoimpeacutedance
acoustique du transducteur et le milieu de propagation Lefficaciteacute de ces couches deacutepend de
la freacutequence et du process Bien que ces capteurs puissent fonctionner dans la gamme de
plusieurs centaines de kHz ils souffrent dune bande de freacutequence eacutetroite et drsquoune sensibiliteacute
relativement faible ce qui entraicircne la faible dynamique du signal
Dautres transducteurs eacutelectro-acoustiques les plus couramment utiliseacutes sont les microphones
de type capacitif et de type pieacutezoreacutesistif Dans le microphone de type capacitif la membrane
fonctionne comme une plaque dun condensateur et les vibrations entraicircnent la variation de la
distance entre les plaques Avec une polarisation DC les plaques stockent une charge fixe
Sous lrsquoeffet de cette charge fixe les surfaces des plaques et le dieacutelectrique au milieu la
tension maintenue agrave travers les plaques de condensateur varie avec la fluctuation de seacuteparation
engendreacutee par la vibration de lair
Le microphone de type pieacutezoreacutesistif est constitueacute dun diaphragme eacutequipeacute de quatre
reacutesistances pieacutezoreacutesistives configureacutees en pont de Wheatstone Les pieacutezoreacutesistances
fonctionnent sur la base de leffet pieacutezoreacutesistif qui deacutecrit la variation de la reacutesistance
eacutelectrique du mateacuteriau agrave cause drsquoune contrainte meacutecanique appliqueacutee Pour les diaphragmes
minces et les petites deacuteformations la variation de la reacutesistance est lineacuteaire en fonction de la
pression appliqueacutee
Le Tableau 1 reacutesume les proprieacuteteacutes de dimensionnement des microphones MEMS de type
capacitif et pieacutezoreacutesistif dans lequel le SBW est deacutefini par le produit de la sensibiliteacute et de la
bande passante du microphone Nous constatons dans le Tableau 1 que en supposant que le
127
ratio daspect du diaphragme reste inchangeacute si les dimensions du microphone sont reacuteduites la
performance globale du microphone pieacutezoreacutesistif va ameacuteliorer tandis que celle du
microphone capacitif deacuteteacuteriore En conseacutequence le meacutecanisme de deacutetection par
pieacutezoreacutesistiviteacute est finalement choisi pour reacutealiser le microphone aeacutero-acoustique
Tableau 1 Proprieacuteteacutes de dimensionnement des microphones MEMS
Microphone type Sensibiliteacute Bande passante SBW Tendance
Piezoreacutesistif 2
2
h
aVB 2
h
a BV
h S minus BW uarr SBW uarr
Capacitif 2
2
h
a
h
A
g
VB 2
h
a
2
2
h
a
g
VB S darr BW uarr SBW darr
Le silicium monocristallin a eacuteteacute principalement utiliseacute pour fabriquer les microphones
aeacutero-acoustiques pieacutezoreacutesistifs gracircce agrave son facteur de jauge tregraves eacuteleveacute Les techniques de
bonding sont utiliseacutees y compris la technique de bonding par fusion agrave haute tempeacuterature et la
technique de bonding direct agrave basse tempeacuterature assisteacute par plasma
Bien que le silicium monocristallin possegravede un facteur de jauge eacuteleveacute le processus de
bonding complique le flux de process et cette technique de bonding noffre pas un rendement
eacuteleveacute Dans les chapitres suivants de cette thegravese le mateacuteriau de silicium polycristallin
re-cristalliseacute sera utiliseacute pour remplacer le silicium monocristallin pour reacutealiser les
pieacutezoreacutesistances
Leacuteleacutement cleacute de la structure de microphone est le film mince qui peut se deacuteformer lorsque la
pression est appliqueacutee Apregraves le processus de deacutepocirct de film mince ce film contient
normalement une contrainte reacutesiduelle qui est le plus souvent provoqueacutee soit par la diffeacuterence
de coefficient de dilatation thermique entre la couche mince et le substrat ou par les
diffeacuterences de proprieacuteteacute de mateacuteriaux dans linterface entre le film mince et le substrat tel que
le deacutesaccord de maille La premiegravere dentre eux est appeleacutee la contrainte thermique et cette
derniegravere est appeleacutee la contrainte intrinsegraveque
128
En 1909 Stoney a constateacute que apregraves le deacutepocirct dun film mince meacutetallique sur le substrat la
structure film-substrat avait plieacute en raison de la contrainte reacutesiduelle dans le film deacuteposeacute
(Figure 1) Puis il a donneacute la formule bien connue comme lEquation 1 pour calculer la
contrainte dans le film mince baseacute sur la mesure de la courbure de flexion du substrat ougrave σ est
la contrainte reacutesiduelle du film mince Es est le module drsquoYoung du mateacuteriau du substrat ds
est lrsquoeacutepaisseur du substrat df repreacutesente leacutepaisseur du film mince νs est le coefficient de
Poisson du mateacuteriau du substrat et R est la courbure de flexion
Figure 1 Flexion de la structure film-substrat en raison de la contrainte reacutesiduelle
fs
ss
dR
dE
)1(6
2
(1)
Le Tableau 2 preacutesente les valeurs numeacuteriques utiliseacutees dans lrsquoEquation 1 pour le calcul et la
contrainte reacutesiduelle calculeacutee
Tableau 2 Les paramegravetres pour la mesure par meacutethode de courbure et le reacutesultat
Es (GPa) νs ds (μm) df (μm) R (m) σ (MPa)
185 028 525 05 1431 165
185 028 525 1 552 214
La formule de Stoney est baseacutee sur lhypothegravese que df ltlt ds et le reacutesultat calculeacute est une
valeur moyenne de la contrainte agrave linteacuterieur du wafer entier La meacutethode de poutre en rotation
est une autre technique couramment utiliseacutee pour mesurer la contrainte reacutesiduelle dans le film
ds Wafer substrate
Thin film
R
df
129
mince et lavantage de cette meacutethode est que la contrainte peut ecirctre mesureacutee localement
Les deacutetails de la structure de poutre en rotation sont preacutesenteacutes dans la Figure 2 Avec les
paramegravetres de conception eacutenumeacutereacutes dans le Tableau 3 leacutequation de calcul de la contrainte
reacutesiduelle est
)(6490
MPaE (2)
ougrave E est le module dYoung du mateacuteriau de la poutre et δ est la distance traverseacute de la poutre
en rotation sous la contrainte Le deacutefaut principal de cette meacutethode est que sauf si nous
savons exactement le module drsquoYoung du mateacuteriau de la poutre la valeur de la contrainte
reacutesiduelle calculeacutee nest pas exacte Les traverseacutees de rotation sont 55μm et 4μm et les
contraintes reacutesiduelles correspondantes sont 175Mpa et 128MPa pour le mateacuteriau LS-SiN
ayant une eacutepaisseur de 1microm et 05microm respectivement Les valeurs de contrainte reacutesiduelle
mesureacutees par la meacutethode de poutre en rotation est denviron 20 de moins que les valeurs
mesureacutees par la meacutethode de courbure
Figure 2 Layout de la structure de poutre en rotation
Wr
Wf
Lf
a
b
h
Lr
130
Tableau 3 Paramegravetres dimensionnelles de la poutre en rotation
Wr (μm) 30 Wf (μm) 30
Lf (μm) 300 Lr (μm) 200
a (μm) 4 b (μm) 75
h (μm) 10
La densiteacute du mateacuteriau du diaphragme et le module drsquoYoung sont eacutegalement importants pour
lestimation de la performance des vibrations meacutecaniques La densiteacute deacutetermine la masse
totale du diaphragme et le module drsquoYoung deacutetermine la constante de raideur du ressort
Toutes ces deux valeurs sont les reacutesultats des calculs indirects de la freacutequence du premier
mode de reacutesonance des structures de poutre encastreacutee-encastreacutee avec des longueurs
diffeacuterentes
LrsquoEacutequation 3 est utiliseacutee pour calculer la freacutequence du premier mode de reacutesonance drsquoune
structure de poutre encastreacutee-encastreacutee baseacute sur la meacutethode de Rayleigh-Ritz ougrave ω est la
freacutequence de reacutesonance en rad s t et L sont respectivement leacutepaisseur et la longueur de la
poutre et E ρ et σ sont respectivement le module drsquoYoung la densiteacute et la contrainte
reacutesiduelle du mateacuteriau de la poutre Comme nous connaissons deacutejagrave la contrainte reacutesiduelle en
utilisant les meacutethodes deacutecrites dans le paragraphe preacuteceacutedent en mesurant les freacutequences du
premier mode de reacutesonance ω1 et ω2 des poutres encastreacutees-encastreacutees avec la mecircme section
transversale mais de diffeacuterentes longueurs L1 et L2 le module dYoung et la densiteacute du
mateacuteriau de la poutre peuvent ecirctre calculeacutes par les Equations 4 et 5
2
2
4
242
3
2
9
4
LL
Et (3)
42
21
22
41
21
22
21
22
2 11
11
2
3
LL
LL
tE
(4)
21
41
22
42
21
22
2
3
2
LL
LL
(5)
131
La freacutequence de reacutesonance de la poutre encastreacutee-encastreacutee est mesureacutee par un vibromegravetre
laser Fogale Le die deacutechantillon est colleacute sur une plaquette pieacutezoeacutelectrique avec silicone
(RHODORSIL trade) et la plaquette pieacutezoeacutelectrique est colleacutee sur une petite carte PCB avec la
colle conductrice dargent Cet eacutechantillon preacutepareacute est fixeacute sur un eacutetage sous vide sans
vibrations Pendant la mesure un signal sinusoiumldal est fourni agrave la plaquette pieacutezoeacutelectrique et
la freacutequence dentreacutee est balayeacutee dans une large bande passante agrave partir de 10kHz jusquagrave 2
MHz Le point laser du vibromegravetre est centreacute au centre de la poutre et lamplitude du
deacuteplacement de la vibration correspondante est enregistreacutee La densiteacute moyenne calculeacutee et le
module drsquoYoung du mateacuteriau LS-SiN deacuteposeacute sont respectivement 3002kgm3 et 207GPa
Pour concevoir un microphone large-bande agrave la haute freacutequence non seulement les
speacutecifications de performance du composant et les proprieacuteteacutes de mateacuteriau doivent ecirctre pris en
compte mais aussi la faisabiliteacute du process de fabrication du composant La conception de la
structure physique doit eacutegalement accompagner la conception du process de fabrication
Les techniques de micro-usinage de surface et de volume ont leurs diffeacuterentes capaciteacutes et
contraintes pour la conception du microphone En utilisant la technique de micro-usinage de
surface les aspects reacutealisables sont les suivants (1) La dimension du diaphragme de deacutetection
suspendu peut ecirctre indeacutependante de leacutepaisseur de la chambre drsquoair au-dessous (2) La
structure concave de pyramide inverseacutee est introduite dans le diaphragme de deacutetection pour
eacuteviter le problegraveme commun de blocage par adheacuterence dans le process de fabrication par
micro-usinage de surface Les limites de cette technique sont les suivantes (1) Les trous
fentes de relaxation seront ouverts sur le diaphragme de deacutetection ce qui conduit agrave un
court-circuit acoustique entre lespace ambiant et la caviteacute en dessous du diaphragme En
raison de ce court-circuit acoustique la diffeacuterence de pression est eacutegaliseacutee agrave basse freacutequence
ce qui limite la performance du microphone (2) A cause de lattaque eacuteventuelle du meacutetal de la
face supeacuterieure par les solutions de gravure la compatibiliteacute du process doit ecirctre prise en
compte
En utilisant la technique de micro-usinage de volume les avantages sont les suivants (1) Il
132
sagit dun process relativement simple et il y a moins de soucis de compatibiliteacute entre la
meacutetallisation en face supeacuterieure et les produits chimiques de relaxation (2) Il y a un
diaphragme complet sans trousfentes ce qui eacutelimine leffet de court-circuit acoustique la
proprieacuteteacute agrave basse freacutequence du microphone sera ainsi ameacutelioreacutee Les contraintes de cette
technique sont les suivantes (1) A cause des caracteacuteristiques de la gravure par cocircteacute infeacuterieur
la caviteacute dair sous le diaphragme de deacutetection sera tregraves large Cela signifie que le diaphragme
de deacutetection ne sera pas amorti Un pic de reacutesonance eacuteleveacute existera dans le spectre freacutequentiel
de la reacuteponse du microphone (2) Quel que soit le type de technique de micro-usinage de
volume utiliseacute la longueur de gravure lateacuterale sera proportionnelle au temps de gravure
verticale Cela signifie que la non-uniformiteacute de leacutepaisseur du substrat conduira agrave une
variation de dimension du diaphragme
Un diaphragme carreacute entiegraverement encastreacute est reacutealiseacute par la technique de micro-usinage de
volume Pour modeacuteliser ses caracteacuteristiques vibratoires la meacutethode drsquoeacuteleacutements finis (FEA)
est la plus approprieacutee Pour un diaphragme carreacute avec une longueur de 210μm agrave laide des
paramegravetres de modeacutelisation eacutenumeacutereacutes dans le Tableau 4 la masse ponctuelle de vibration est
simuleacutee agrave 39510-11 kg et la freacutequence du premier mode de reacutesonance est drsquoenviron 840kHz
Tableau 4 Paramegravetres de modeacutelisation du diaphragme carreacute
Longueur du diaphragme (m) 210 Epaisseur du diaphragme (m) 05
Densiteacute du diaphragme (SiN)
(kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
En utilisant lrsquoeacutequation suivante
m
kfr 2
1 (6)
ougrave fr est la freacutequence de reacutesonance k est la constante effective de ressort du diaphragme et m
est la masse effective du diaphragme le k est calculeacute drsquoecirctre 1100Nm La reacuteponse
freacutequentielle meacutecanique du diaphragme peut ecirctre modeacuteliseacutee par un simple systegraveme de
133
ressort-masse avec un seul degreacute de liberteacute Ensuite en utilisant lrsquoanalogie eacutelectro-meacutecanique
la reacuteponse freacutequentielle meacutecanique du capteur peut ecirctre analyseacutee en utilisant la theacuteorie
traditionnelle du circuit eacutelectrique
Compte tenu de la technique de micro-usinage de surface en raison de la fente requise pour la
gravure de relaxation la structure drsquoun diaphragme carreacute avec quatre poutres de support est
utiliseacutee et la fente de gravure entoure les poutres de support et le diaphragme (Figure 3)
Comme deacutecrit dans la section preacuteceacutedente en raison de lrsquoeffet de court-circuit acoustique
introduit par la fente de relaxation il est difficile de modeacuteliser analytiquement la reacuteponse
coupleacutee acoustique-meacutecanique Dans cette situation seule la meacutethode par eacuteleacutements finis est
applicable agrave la modeacutelisation de cet effet compliqueacute Sous ANSYS lrsquoeacuteleacutement 3-D acoustique
de la fluide FLUID30 est utiliseacute pour modeacuteliser le milieu fluide lair dans notre cas et
linterface dans les problegravemes dinteraction fluide-structure Lrsquoeacuteleacutement infini 3-D acoustique
de la fluide FLUID130 est utiliseacute pour simuler les effets dabsorption drsquoun domaine de fluide
qui seacutetend agrave linfini au-delagrave de la limite du domaine constitueacute des eacuteleacutements FLUID30 Le
20-noeud eacuteleacutement structurel solide SOLID186 est utiliseacute pour modeacuteliser la deacuteformation
meacutecanique de la structure et les proprieacuteteacutes de la vibration
Figure 3 Layout du diaphragme avec les poutres de support (les reacutesistances de reacutefeacuterence ne sont pas indiqueacutees)
a
l
w
Heavily doped area
Sensing area
Sensing diaphragm
Releasing slot
134
Les paramegravetres de modeacutelisation sont eacutenumeacutereacutes dans le Tableau 5 Lrsquoanalyse harmonique est
appliqueacutee sur le modegravele en balayant la freacutequence de 10Hz agrave 1MHz La simulation de la
reacuteponse freacutequentielle meacutecanique montre que la freacutequence du premier mode de reacutesonance est
400kHz
Tableau 5 Paramegravetres de modeacutelisation en couplage acoustique-meacutecanique
Longueur du diaphragme
(μm)
115 Epaisseur du diaphragme (μm) 05
Longueur du diaphragme
de support (μm)
55 Largeur du diaphragme de
support (μm)
25
Profondeur de la caviteacute
drsquoair (μm)
9 Rayon de la plaque
drsquoabsorption acoustique (μm)
345
Longueur de la fente de
relaxation (μm)
700 Largeur de la fente de
relaxation (μm)
5
Densiteacute du diaphragme
(SiN) (kgm3)
3002 Module drsquoYoung du
diaphragme (SiN) (GPa)
207
Coefficient de Poisson 027 Contrainte reacutesiduelle (MPa) 165
Vitesse de son (ms) 340 Densiteacute drsquoair (kgm3) 1225
Le silicium monocristallin (sc-Si) est un mateacuteriau tregraves mature dans lindustrie des
semiconducteurs pour les applications pieacutezoreacutesistives Toutefois agrave cause des limitations du
mateacuteriau et de la technologie tels que le deacutesaccord de maille les diffeacuterents coefficients de
dilatation thermique et le rendement de bonding il est assez difficile et coucircteux drsquointeacutegrer le
mateacuteriau sc-Si sur les substrats exotiques comme le verre pour les applications daffichage sur
le panneau plat ou de lrsquointeacutegrer dans les circuits inteacutegreacutes en 3-D comme dans un process de
fabrication du VLSI
Au lieu de sc-Si lrsquoa-Si fabriqueacute par les techniques de deacutepocirct LPCVD ou PECVD est utiliseacute
pour fabriquer les circuits de pilotage avec les transistors agrave couche mince (TFT) pour les
eacutecrans agrave cristaux liquides et les cellules photovoltaiumlques inteacutegreacutees sur les substrats en verre ou
135
en plastique Lrsquoinconveacutenient principal du mateacuteriau a-Si est sa faible mobiliteacute agrave effet de champ
Par conseacutequent la technique de deacutepocirct du silicium cristallin sur les mateacuteriaux amorphes
devient de plus en plus importante pour lindustrie des semiconducteurs et la taille des grains
du silicium deacuteposeacute est une consideacuteration particuliegraverement pertinente pour le process puisque
la taille du grain peut dominer les proprieacuteteacutes eacutelectriques des mateacuteriaux qui ont une faible taille
du grain
Entre sc-Si et a-Si le poly-Si est composeacute de petits cristaux appeleacutes cristallites Il est
consideacutereacute comme un mateacuteriau preacutefeacutereacute par rapport agrave a-Si en raison de sa mobiliteacute de porteuses
bien plus eacuteleveacutee Le mateacuteriau poly-Si peut ecirctre directement deacuteposeacute dans un four LPCVD sur
une plate-forme de PECVD ou cristalliseacute agrave lrsquoissu de la-Si deacuteposeacute par les mecircmes techniques
mentionneacutees ci-dessus La qualiteacute des films minces de poly-Si cristalliseacute a un effet important
sur la performance des dispositifs de poly-Si Au cours des deux derniegraveres deacutecennies de
diffeacuterentes technologies ont eacuteteacute proposeacutees pour la cristallisation de la-Si sur les substrats
exotiques y compris la cristallisation en phase solide (SPC) la cristallisation par laser agrave
excimegravere (ELC) et la cristallisation par lrsquoinduction lateacuterale meacutetallique (MILC)
Dans le processus de SPC le recuit thermique fournit leacutenergie neacutecessaire agrave la nucleacuteation et
lrsquoexpansion des grains En geacuteneacuteral la cristallisation intrinsegraveque en phase solide a besoin dune
longue dureacutee pour cristalliser complegravetement lrsquoa-Si en tempeacuterature eacuteleveacutee et une grande
densiteacute de deacutefaut existe toujours dans le poly-Si cristalliseacute
La cristallisation par laser est une autre meacutethode largement utiliseacutee dans lrsquoactualiteacute pour
preacuteparer le poly-Si sur les substrats exotiques En geacuteneacuteral le processus ELC est capable de
produire les mateacuteriaux de haute qualiteacute mais elle souffre dun faible rendement et drsquoun coucirct
eacuteleveacute des eacutequipements
Afin de maintenir agrave la fois un rendement eacuteleveacute et une grande taille du grain le proceacutedeacute de
cristallisation assisteacutee par les catalyseurs a eacuteteacute inventeacute et peut ecirctre diviseacute en deux grandes
cateacutegories ceux utilisant les catalyseurs semiconducteurs comme le germanium et ceux
utilisant les meacutetaux tels que Al Au Ag Pd Co et Ni Les meacutetaux sont drsquoabord deacuteposeacutes sur
136
la-Si Puis la-Si est cristalliseacute en polysilicium agrave une tempeacuterature infeacuterieure agrave celle du CPS
Reacutecemment le Ni MILC a attireacute beaucoup dattention Le preacutecipiteacute NiSi2 joue le rocircle drsquoun
noyau de silicium qui preacutesente une structure cristalline semblable au silicium avec un
deacutesaccord de maille de 04 avec le silicium La constante de reacuteseau de NiSi2 5406Aring est
presque eacutegale agrave celle du silicium 5430Aring
Le process complet de micro-usinage de surface commence agrave partir dun wafer de silicium de
type p (100) La premiegravere eacutetape consiste agrave former les moules concaves des pyramides inverses
en surface du substrat La deuxiegraveme eacutetape est de deacuteposer et structurer la couche sacrificielle
Apregraves le deacutepocirct des couches sacrificielles un masque de laquotrancheacuteeraquo est utiliseacute pour faire la
photolithographie pour structurer la zone du diaphragme Ensuite la-Si est graveacute par LAM
490 Enfin loxyde humide est graveacute par la solution BOE Apregraves lrsquoenlegravevement de la reacutesine le
LS-SiN de 400nm est deacuteposeacute par LPCVD suivi dune couche de silicium agrave 600nm Ensuite
un masque de reacutesistances est utiliseacute pour la photolithographie et la-Si est graveacute par un plasma
agrave couplage inductif (ICP) agrave laide du gaz HBr Leacutetape suivante est de cristalliser la-Si en
poly-Si en utilisant la technique MILC Tout dabord un oxyde de basse tempeacuterature (LTO) de
300nm est deacuteposeacute Ensuite un masque pour le trou de contact est utiliseacute pour la
photolithographie et le BOE est utiliseacute pour graver le LTO Apregraves lrsquoenlegravevement de la reacutesine et
une plongeacutee dans la solution HF (1100) une couche de nickel de 5 nm est eacutevaporeacutee sur la
surface du wafer A lrsquoissu drsquoun recuit dans latmosphegravere dazote agrave 590degC pendant 24 heures le
mateacuteriau amorphe est induit au type polycristallin Ensuite le nickel est enleveacute par une
solution piranha qui est suivi dun recuit agrave haute tempeacuterature agrave 900degC pendant 05 heure
Apregraves la cristallisation de la-Si la solution BOE est utiliseacutee pour deacutecaper la couche LTO et le
bore est dopeacute dans le mateacuteriau poly-Si par technique dimplantation ionique Apregraves cela les
eacutechantillons sont mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant Par la
suite en deacuteposant la deuxiegraveme couche de LS-SiN (100nm) les pieacutezoreacutesistances en poly-Si
sont bien proteacutegeacutes Ensuite un masque de trou de contact et la reacutesine FH 6400L sont utiliseacutes
pour faire la photolithographie et le RIE 8110 est effectueacute pour graver le nitrure de silicium
Un fort dopage au bore est reacutealiseacute ici pour reacuteduire la reacutesistance de contact Agrave la suite de
137
limplantation dimpureteacute une activation est effectueacutee agrave 900degC pendant 30 minutes Apregraves
avoir utiliseacute la technique de siliciure de titane pour ameacuteliorer la reacutesistance de contact le
masque de trou de gravure est utiliseacute pour faire la photolithographie et le RIE 8110 est
effectueacute pour graver le nitrure de silicium et ouvrir les trous de gravure de relaxation Le
systegraveme de meacutetallisation drsquoune double-couche de chrome et dor est deacuteposeacute par un proceacutedeacute de
lift-off Apregraves le lift-off les lignes meacutetalliques sont bien deacutefinies La derniegravere eacutetape consiste agrave
deacutecaper les deux couches sacrificielles (y compris loxyde et la-Si) et agrave libeacuterer le diaphragme
La figure 4 preacutesente un microphone large-bande agrave haute freacutequence fabriqueacute avec succegraves en
utilisant la technique de micro-usinage de surface
Figure 4 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de surface
La technique de micro-usinage de volume commence par un wafer poli de type p (100) avec
une eacutepaisseur de 300microm Au deacutebut une couche doxyde thermique de 05microm une couche
drsquoa-silicium de 01microm et une couche de LS-SiN de 04μm en eacutepaisseur sont deacuteposeacutees dans
lordre Ensuite une couche de lrsquoa-Si de 0s6μm en eacutepaisseur est deacuteposeacutee en tant que mateacuteriau
pieacutezoreacutesistif Le mateacuteriau drsquoa-silicium en face infeacuterieure est enleveacute par une machine de
gravure LAM 490 et la face supeacuterieure du silicium est cristalliseacutee en poly-Si en utilisant la
technique MILC Cette couche cristalliseacutee de silicium polycristallin est ensuite structureacutee pour
former la forme des pieacutezoreacutesistances La pieacutezoreacutesistance est dopeacutee au bore par implantation
Apregraves cela le wafer est mis dans le four agrave 1000degC pendant 15 heure pour activer le dopant
Apregraves lactivation du dopant une deuxiegraveme couche de LS-SiN de 01microm en eacutepaisseur est
Sensing
diaphragm
Reference resistor
Sensing resistor
115μm
138
deacuteposeacutee puis une couche de LTO de 2microm drsquoeacutepaisseur est deacuteposeacutee et le mateacuteriau LTO de la
face supeacuterieure est enleveacute par la solution BOE Ensuite le trou de contact est ouvert agrave laide
de la reacutesine FH 6400L et la machine agrave graver agrave sec RIE 8110 La zone de contact est
fortement dopeacutee au bore Agrave la suite de limplantation dimpureteacute lrsquoactivation est effectueacutee agrave
900degC pendant 30 minutes Apregraves cela une couche drsquoAl Si drsquoune eacutepaisseur de 05microm est
pulveacuteriseacutee et structureacutee pour former la meacutetallisation Un recuit dans le gaz drsquoazote hydrogeacuteneacute
agrave 400degC pendant 30 minutes est effectueacute pour ameacuteliorer la reacutesistiviteacute de contact Ensuite une
reacutesine eacutepaisse de 3microm PR507 est deacuteposeacutee sur la face infeacuterieure du wafer et structureacutee pour
former la zone du diaphragme Les mateacuteriaux deacuteposeacutes en face infeacuterieure comme LTO LS-SiN
a-Si et loxyde thermique sont enleveacutes par la technique de gravure segraveche Apregraves cela le
substrat en silicium est graveacute agrave travers par la technique DRIE Puis loxyde et lrsquoa-silicium du
cocircteacute supeacuterieur sont eacutegalement eacutelimineacutes en utilisant la technique de gravure segraveche La figure 5
preacutesente un microphone fabriqueacute en utilisant la technique de micro-usinage de volume
Figure 5 Microphotographe drsquoun microphone large-bande agrave haute freacutequence fabriqueacute en utilisant la technique de micro-usinage de volume
Apregraves la fabrication du dispositif la reacutesistance carreacutee du mateacuteriau poly-Si MILC dopeacute est
mesureacutee par une structure en croix grecque Pour leacutechantillon fabriqueacute par la technique de
micro-usinage de surface les reacutesistances carreacutees en moyenne mesureacutees dans la zone de
deacutetection et dans la zone de connexion sont respectivement 4114 ohmscarreacute (Ω) et
247Ω Pour leacutechantillon fabriqueacute par la technique de micro-usinage de volume la
Sensing
diaphragm
Sensing resistor
Reference resistor
210μm
139
reacutesistance carreacutee en moyenne mesureacutee dans la zone de deacutetection est 4464Ω Comme les
reacutesistances de deacutetection sont fabriqueacutees en utilisant la mecircme technique MILC avec le mecircme
dopage drsquoimpureteacute et la mecircme condition dactivation pour les deux techniques leurs
reacutesistances carreacutees sont presque identiques
La structure de Kelvin est utiliseacutee pour mesurer la reacutesistance de contact Rc entre le meacutetal et le
mateacuteriau poly-Si MILC dopeacute Pour le systegraveme de contact entre CrAu et poly-Si MILC la
reacutesistance de contact en moyenne est mesureacutee agrave 466Ω et la reacutesistiviteacute speacutecifique de contact
est 291μΩbullcm2 (avec une surface de contact de 625μm2) et pour le systegraveme de contact entre
Al Si et poly- Si MILC la reacutesistance de contact en moyenne est mesureacutee agrave 58Ω et la
reacutesistiviteacute speacutecifique de contact est 232μΩbullcm2 (avec une surface de contact de 4μm2) Avec
laide de la couche auto-aligneacutee de siliciure de titane la reacutesistiviteacute speacutecifique de contact du
systegraveme CrAu et poly-Si MILC est seulement leacutegegraverement plus grande que celle du systegraveme
traditionnel Al Si et poly-Si MILC
La configuration de mesure statique est illustreacutee dans la Figure 6 La puce fabriqueacutee est lieacutee
par fil sur une carte PCB Cette derniegravere est ensuite colleacutee sur un support meacutetallique et fixeacute
sur une platine exempt de vibrations Un tribo-indenteur controcircleacute par ordinateur est utiliseacute
pour appliquer un point de charge au centre du diaphragme de deacutetection Un pont de
Wheatstone composeacute de deux reacutesistances de deacutetection et deux de reacutefeacuterence respectivement
sur et hors de la membrane est utiliseacute pour mesurer la reacuteponse statique agrave la force dans le
diaphragme Avec une polarisation DC dentreacutee la tension en sortie est mesureacutee et enregistreacutee
en utilisant un analyseur de paramegravetre du semiconducteur HP 4155 Pour le diaphragme de
115x115μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de surface avec une
polarisation DC de 2V une reacuteponse statique drsquoenviron 04μVVPa est mesureacutee Et pour le
diaphragme de 210x210μm2 carreacute qui est fabriqueacute par la technique de micro-usinage de
volume avec une polarisation DC de 3V une reacuteponse statique drsquoenviron 028μVVPa est
mesureacutee
140
Figure 6 Configuration de la mesure statique
La reacuteponse dynamique du microphone est mesureacutee par une source acoustique lrsquoonde N La
meacutethode la plus courante de geacuteneacuteration de lrsquoonde N est la stimulation dune eacutetincelle
eacutelectrique agrave haute tension Un circuit simple de deacutecharge drsquoeacutetincelle est illustreacute dans la Figure
7 Une alimentation agrave haute tension (~14kV) charge un condensateur de stockage (1nF) agrave
travers une reacutesistance de limitation de courant (50M) et la deacutecharge du condensateur se
produit agrave travers lespace de deacutecharge (~ 13cm)
Figure 7 Scheacutema du condensateur de deacutecharge agrave haute tension
Comme nous lrsquoavons trouveacute dans la mesure statique de nano-indentation la sensibiliteacute de
leacutechantillon est tregraves faible Ainsi une carte damplification est rajouteacutee agrave la sortie du capteur
pour augmenter le signal et le rendre suffisamment grand pour ecirctre captureacute par loscilloscope
PC controller
Triboindentor
(Hysitron)
Sample
Stage
~14kV
1nF
50MΩ
Spark gap ~13cm
141
Apregraves avoir deacutecouvert lexacte forme de lrsquoonde N agrave une distance r0 de la source deacutetincelle
nos eacutechantillons agrave calibrer sont placeacutes agrave cette mecircme distance Un signal typique de lrsquoonde N
mesureacute sur les dispositifs de micro-usinage de surface est preacutesenteacute dans la Figure 8 Sur la
figure on peut clairement identifier deux signaux conseacutecutifs doscillation La premiegravere
oscillation correspond agrave la forte hausse du choc drsquoavant de lrsquoonde N et la seconde oscillation
correspond agrave la forte hausse du choc drsquoarriegravere de lrsquoonde N Toutefois les informations de
basse freacutequence de lrsquoonde N correspondant agrave la pente de lavant vers lrsquoarriegravere du choc ne sont
pas visibles dans la courbe mesureacutee Cela permet aussi de veacuterifier la perte dinformation agrave
basse freacutequence ducirc agrave leffet de court-circuit acoustique qui est preacutevu dans la modeacutelisation par
eacuteleacutements finis
La reacuteponse en freacutequence du microphone calibreacute est preacutesenteacutee dans la Figure 9 qui est
eacutegalement compareacutee avec le reacutesultat FEA Le pic de reacutesonance est denviron 400kHz qui est
eacutegal agrave la preacutediction du reacutesultat FEA La bande plate est tregraves eacutetroite agrave peu pregraves de 100kHz agrave
200kHz et au-dessous de 100kHz la reacuteponse en freacutequence est rapidement diminueacutee La
sensibiliteacute dynamique agrave linteacuterieur de la bande plate est 0033μVVPa qui est bien plus faible
que la valeur statique (04μVVPa) Ce pheacutenomegravene peut aussi sexpliquer par leffet de
court-circuit acoustique Mecircme si on peut trouver exactement la pression incidente P0 au
diaphragme la diffeacuterence reacuteelle de pression ∆p sur le diaphragme de deacutetection est eacutegale agrave P0 -
Ps (Ps est la pression de fuite dans la caviteacute dair agrave travers les trous fentes de relaxation) qui
par la technique de micro-usinage de surface (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 9 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et le signal en moyenne) en comparaison avec le reacutesultat du FEA
La Figure 10 montre le signal typique drsquoonde N mesureacute en utilisant les dispositifs de
micro-usinage de volume Dapregraves la Figure 11 il est montreacute que les dispositifs micro-usineacutes
en volume ont une freacutequence de reacutesonance plus haute (715kHz) et dans la Figure 10 on peut
voir que non seulement les informations agrave haute freacutequence mais aussi les informations agrave
basse freacutequence peuvent ecirctre prises par ce dispositif En outre nous pouvons constater quil y
a une oscillation superposeacutee sur la pente ce qui signifie que le dispositif microphone nest pas
suffisamment amorti agrave sa freacutequence de reacutesonance
143
Figure 10 Reacutesultat typique drsquoune mesure agrave eacutetincelle drsquoun eacutechantillon de microphone fabriqueacute par la technique de micro-usinage de volume (Polarisation DC 3V avec le gain drsquoamplification agrave 1000 et la distance entre la source et le microphone agrave 10cm)
Figure 11 Spectre drsquoamplitude du FFT unilateacuteral des signaux mesureacutes par un microphone micro-usineacute en volume et par la meacutethode optique
La reacuteponse en freacutequence des dispositifs de micro-usinage de volume est preacutesenteacutee dans la
Figure 12 qui est compareacute avec le reacutesultat de modeacutelisation par eacuteleacutements concentreacutes La
sensibiliteacute dynamique est 1mVPa (avec un gain damplification de 1000 et une polarisation
DC de 3V) ce qui signifie que la sensibiliteacute dynamique reacuteelle du microphone est denviron
033μVVPa et similaire agrave la sensibiliteacute statique calibreacutee (028μVVPa) En outre ce
microphone preacutesente une bande passante large et plate de 6kHz agrave 500kHz
fr = 715kHz
144
Figure 12 La reacuteponse en freacutequence du microphone calibreacute (Polarisation DC 3V avec le gain drsquoamplification de 1000 et le signal en moyenne) en comparaison avec le reacutesultat de
modeacutelisation par eacuteleacutements concentreacutes
Enfin trois capteurs micro-usineacutes en volume sont placeacutes dans un plan pour former un reacuteseau
qui deacutemontre son lapplication comme un localisateur sonore (Figure 13) Le premier capteur
(M1) preacutesente une coordonneacutee de x1 = 25 y1 = 0 et z1 = 0 le deuxiegraveme capteur (M2) preacutesente
une coordonneacutee de x2 = -25 y2 = 0 et z2 = 0 et le troisiegraveme capteur (M3) preacutesente une
coordonneacutee de x3 = 0 y3 = 4 et z3 = 0 avec toutes les uniteacutes en centimegravetre
Figure 13 Coordonneacutees du reacuteseau de capteurs
La Figure 14 preacutesente la configuration de lapplication de localisation de la source sonore Le
geacuteneacuterateur deacutetincelles eacutemit londe acoustique qui est deacutetecteacutee par le reacuteseau de capteurs Les
signaux deacutetecteacutes sont captureacutes par un oscilloscope (Tektronix TDS 2024C) puis les signaux
M1 M2
M3
X
Y
0
145
captureacutes sont transfeacutereacutes agrave un ordinateur portable via un cacircble USB en utilisant le MATLAB
Instrument Control Toolbox qui est baseacute sur le standard NI-VISA de National Instruments
Ensuite les temps de deacutelai et les coordonneacutees de source acoustique sont calculeacutes par le logiciel
MATLAB Toutes ces fonctions sont reacutealiseacutees par une interface utilisateur graphique (GUI)
personnaliseacutee sous MATLAB
Figure 14 La configuration du systegraveme de localisation de la source acoustique
La source sonore drsquoeacutetincelle est preacutereacutegleacutee aux coordonneacutees (xs = 0cm ys = 4cm) dans le plan
XY Etant donneacute que les deux aiguilles deacutetincelle ont un eacutecart de 13 cm entre eux la position
meacutediane entre les aiguilles est supposeacutee ecirctre la position de la source (Figure 15) La distance
entre la source sonore et le reacuteseau de capteurs en coordonneacutee Z varie de 10cm agrave 105cm (la
distance est mesureacutee par une regravegle) Agrave chaque position 20 mesures sont effectueacutees En
utilisant les temps de deacutelai mesureacutes et la vitesse du son calibreacute les coordonneacutees de la source
sonore sont calculeacutees et compareacutees avec les valeurs qui ont eacuteteacute mesureacutees au preacutealable par la
regravegle (Figure 16)
Figure 15 Deacutefinition de la position de la source sonore
0 Z
(xs = 0cm ys = 4cm )
Spark needle Spark needle
13cm X
Assumed source position
Y
146
Figure 16 Les comparaisons de coordonneacutees entre les valeurs preacute-mesureacutees et les valeurs
calculeacutees sur (a) les coordonneacutees X (b) les coordonneacutees Y (c) les coordonneacutees Z
Dapregraves la Figure 16 il est montreacute que les valeurs preacute-mesureacutees et les valeurs calculeacutees des
coordonneacutees Z correspondent tregraves bien contrairement aux coordonneacutees X et Y Pour les
coordonneacutees X (Figure 16 (a)) les valeurs calculeacutees fluctuent autour des valeurs preacute-mesureacutees
Ce pheacutenomegravene peut ecirctre expliqueacute par le fait que le point reacuteel de geacuteneacuteration deacutetincelle nrsquoest
(c)
(b)
(a)
147
pas toujours au milieu des deux aiguilles Au contraire ce point oscille au cours de
lexpeacuterience aux diffeacuterentes positions Pour veacuterifier cette hypothegravese une cameacutera agrave haute
vitesse est neacutecessaire pour capturer les images deacutetincelle lors des mesures pour lanalyse de
position ce qui nest pas applicable au stade actuel
Pour les coordonneacutees Y (Figure 16 (b)) les diffeacuterences entre les valeurs preacute-mesureacutees et les
valeurs calculeacutees augmentent lineacuteairement jusquagrave 2cm lorsque la position de mesure passe de
1 agrave 11 (dans la figure 16 (c) cela signifie que la coordonneacutee Z varie de 10cm agrave 105cm) Les
diffeacuterences entre les valeurs preacute-mesureacutees et les valeurs calculeacutees des coordonneacutees Y peuvent
sexpliquer par la deacutenivellation de la surface du sol Langle θ entre la surface du sol et le
niveau est calculeacute agrave 11deg
Bien que les deux prototypes de microphone large-bande agrave haute freacutequence sont fabriqueacutes
avec succegraves et calibreacutes il y a plusieurs sujets agrave reacutesoudre en avenir Tout dabord les modegraveles
de ces deux microphones sont tous baseacutes sur la meacutethode FEA Un modegravele analytique qui
nrsquoest pas tregraves preacutecis mais pourrait estimer rapidement la performance des diffeacuterentes
structures est bien neacutecessaire Dautre part concernant le microphone fabriqueacute par la
technique de micro-usinage de volume en raison de la grande caviteacute sous le diaphragme de
deacutetection il ny a pas damortissement suffisant pour amortir le critique pic de reacutesonance
Dans le futur une nouvelle structure avec un amortisseur inteacutegreacute utilisant le lrsquoeffet
damortissement du film de compression doit ecirctre exploreacutee Troisiegravemement lors de nos tests
lamplificateur est reacutealiseacute par les composant discrets sur le PCB et relieacute au capteur par wire
bonding Afin drsquoatteacutenuer le bruit et drsquoameacuteliorer la performance damplification lamplificateur
doit ecirctre fabriqueacute sur la mecircme puce que le capteur et eacuteventuellement le capteur et
lamplificateur peuvent ecirctre fabriqueacutes sur le mecircme substrat
Titre Microcapteurs de Hautes Freacutequences pour des Mesures en Aeacuteroacoustique Reacutesumeacute
Lrsquoaeacuteroacoustique est une filiegravere de lacoustique qui eacutetudie la geacuteneacuteration de bruit par un mouvement fluidique turbulent ou par les forces aeacuterodynamiques qui interagissent avec les surfaces Ce secteur en pleine croissance a attireacute des inteacuterecircts reacutecents en raison de lrsquoeacutevolution de la transportation aeacuterienne terrestre et spatiale Les microphones avec une bande passante de plusieurs centaines de kHz et une plage dynamique couvrant de 40Pa agrave 4 kPa sont neacutecessaires pour les mesures aeacuteroacoustiques Dans cette thegravese deux microphones MEMS de type pieacutezoreacutesistif agrave base de silicium polycristallin (poly-Si) lateacuteralement cristalliseacute par lrsquoinduction meacutetallique (MILC) sont conccedilus et fabriqueacutes en utilisant respectivement les techniques de microfabrication de surface et de volume Ces microphones sont calibreacutes agrave laide dune source drsquoonde de choc (N-wave) geacuteneacutereacutee par une eacutetincelle eacutelectrique Pour leacutechantillon fabriqueacute par le micro-usinage de surface la sensibiliteacute statique mesureacutee est 04microVVPa la sensibiliteacute dynamique est 0033microVVPa et la plage freacutequentielle couvre agrave partir de 100 kHz avec une freacutequence du premier mode de reacutesonance agrave 400kHz Pour leacutechantillon fabriqueacute par le micro-usinage de volume la sensibiliteacute statique mesureacutee est 028microVVPa la sensibiliteacute dynamique est 033microVVPa et la plage freacutequentielle couvre agrave partir de 6 kHz avec une freacutequence du premier mode de reacutesonance agrave 715kHz
Mots-Cleacutes Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Title High Frequency MEMS Sensor for Aero-acoustic Measurements Abstract
Aero-acoustics a branch of acoustics which studies noise generation via either turbulent fluid motion or aerodynamic forces interacting with surfaces is a growing area and has received fresh emphasis due to advances in air ground and space transportation Microphones with a bandwidth of several hundreds of kHz and a dynamic range covering 40Pa to 4kPa are needed for aero-acoustic measurements In this thesis two metal-induced-lateral-crystallized (MILC) polycrystalline silicon (poly-Si) based piezoresistive type MEMS microphones are designed and fabricated using surface micromachining and bulk micromachining techniques respectively These microphones are calibrated using an electrical spark generated shockwave (N-wave) source For the surface micromachined sample the measured static sensitivity is 04microVVPa dynamic sensitivity is 0033microVVPa and the frequency range starts from 100kHz with a first mode resonant frequency of 400kHz For the bulk micromachined sample the measured static sensitivity is 028microVVPa dynamic sensitivity is 033microVVPa and the frequency range starts from 6kHz with a first mode resonant frequency of 715kHz
Keywords Aero-acoustic Bulk micro-machining DRIE MEMS Microphone MILC Wide-band
Laboratoire TIMA ndash 46 avenue Feacutelix Viallet 38000 Grenoble France ISBN 978-2-11-129173-7